SF
StudyFix / Manom's Playbook
Handover · v2.0 · For Manom

You don't have to pick a job. You have to pick a lane.

This isn't a "here's your role" doc. It's a menu. Everything a small business needs someone to own is on it — with honest notes on who thrives in each seat, who gets crushed by it, and how it actually shows up at StudyFix. Browse it. Star a few. Tell me at the end of the week what feels like you.

Sections 2–5 are curriculums — how to actually learn the four backend lanes you're curious about. Each is a 5-week week-by-week where every week ends in you having done something, not just read something. Pick one, or run two in parallel.

Device: iPad + Magic Keyboard Sources: Hormozi playbooks · Leila's advice · Acquisition.com Updated: 02 Jul 2026
◆ Before you start

The AI toolkit — 5-minute crash course

Every one of the four lanes below becomes 10× easier when you know which AI to use for what. Here's the map. Learn the basics; each curriculum below ends with the specific AI workflow for that lane.

Claude (Anthropic)
Pro $20 · Max $100–200/mo
StudyFix's default. Best for reasoning, long documents, code review, prompt-eng iteration, medical writing. Claude Code is the CLI you use for actual dev work. Projects let you drop reference docs once and reuse across chats.
Gemini (Google)
Advanced $20 · Ultra $249/mo
Best for research. "Deep Research" mode auto-crawls dozens of sources. Extremely long context — feed it a whole textbook chapter. Gems = Google's version of Custom GPTs. Weaker at writing than Claude.
ChatGPT (OpenAI)
Plus $20 · Pro $200/mo
Best for images, voice, spreadsheets. DALL-E for thumbnails, Sora for short video, voice-mode for rubber-ducking, Code Interpreter for CSV analysis. Custom GPTs if you want a reusable persona.
Utility layer
$0–$30/mo each
Perplexity for real-time web search. Cursor for AI-in-your-editor coding. v0.dev for UI mockups from a text prompt. Descript for AI-native video editing (auto filler removal).

What makes AI output good vs bad — 5 rules that apply to every lane

  1. Give context, not just questions. The AI can't infer StudyFix voice from nothing. Paste the memory file, the style guide, the prior example.
  2. Give examples over instructions. "Rewrite this the way @personalbrandlaunch sounds" beats "make it more casual." Show, don't tell.
  3. Constrain the output. Word count, format, banned words, required structure. Vague prompts get vague answers.
  4. Iterate — don't accept first draft. Reject specifically ("the hook is too long, cut to 6 words"), don't restart the whole chat.
  5. Verify facts, especially medical ones. AI will confidently invent a drug dose. Every claim gets cross-checked against BNF / NICE / source PDF before it ships.

reality check — what works, what needs a workaround

You're on iPad + Magic Keyboard. Most of this doc works natively. The exceptions all live in the coding lane — here's the map.

Tool mentioned iPad Best iPad workaround
Claude Code CLI Not native Claude.ai web + Projects in Safari covers 95% of the same power (Artifacts, file uploads, project instructions). For real terminal work → SSH into a Mac or cloud VM via Blink Shell (~$20 one-time).
Cursor IDE Not native GitHub Codespaces in Safari — full VS Code in-browser. Click Code → Codespaces on any StudyFix repo. Free tier is generous; upgrade if you outgrow it.
Local dev server (pnpm dev) Not native Codespaces port forwarding — dev server runs on the cloud VM, previews in a Safari tab with a private URL. Same experience as localhost.
Terminal / bash / node / npm Limited Blink Shell (best, ~$20 one-time) for full terminal + SSH. a-Shell (free) for a lighter baseline. Working Copy ($20 pro) for git-focused workflows.
Everything else in this doc
Claude / Gemini / ChatGPT / CapCut / Later / Metricool / Notion / Stripe / Wise / Descript / v0.dev / Anthropic Console / GitHub web		 Native All have iPad apps or work in Safari. Zero workarounds needed. This is why iPad-only actually works for your role.
01 /

Pick a lane — every possible seat in the business

Grouped by Leila Hormozi's five pillars — Marketing, Sales, Product, People, Profit. Tap a pillar chip to filter. Every role card names the Hormozi playbook that goes deep on it, so if a role interests you, you know exactly which PDF to open next.

Not sure where to look first? Tap every option that fits.

Pick as many as feel true — you're allowed to want more than one thing. The bars at the end tally which pillars kept showing up in your answers.

1. Where does your energy come from at the end of a long day?
2. A project stalls. Instinct?
3. Which sentence do you want written on your gravestone?
Your pillar mix

Ranked by how often each pillar showed up. Start with the top one — but the second and third are worth a proper look too if they're close.

0
Your shortlist

Tap + on any role to add it. Aim for 3 to try for real this week.

Marketing Creative

Head of Content / Creator Ops

The face & hands of the brand on IG, TikTok, YouTube.

Day-to-dayBatch-films, edits Reels, writes hooks, replies to comments, watches watch-time analytics. Owns the content calendar.
Thrives if youenjoy being on camera, love a great hook, don't mind shipping imperfect & iterating.
Drains you if youhate the sound of your own voice, need long feedback loops, get anxious about public metrics.
StudyFix fitHighest leverage lane. Youhana already does long-form; the short-form Reel/TikTok/Shorts side is wide open and urgent.
Source · $100M Marketing Machine · $100M Hooks Playbook
Marketing Analytical

Media Buyer / Paid Ads

Pays Meta & TikTok to put us in front of med students.

Day-to-dayAds Manager, targeting, creative testing, CPA tracking. Kills losing ads early, scales winners.
Thrives if youlike spreadsheets, treat every ad as a hypothesis, aren't emotionally attached to your own creative.
Drains you if youget bored by dashboards, take a losing ad personally, hate calling losers early.
StudyFix fitNot yet — we don't run paid until organic proves the offer converts. Bookmark for 2027.
Source · $100M Goated Ads Playbook
Marketing Creative

Brand & Voice Lead

Owns how StudyFix sounds, looks, and feels everywhere.

Day-to-dayGuards the anti-AI-slop voice. Signs off on visuals, copy, launch narratives. Says "no" to anything that sounds like a wellness brand.
Thrives if youhave strong taste and aren't afraid to overrule, notice when something feels 5% off, love reading great copy.
Drains you if youare conflict-averse, want a checklist, hate re-reviewing the same landing page copy for the 4th time.
StudyFix fitBig open seat. The brand's voice is set — but nobody enforces it. Content already drifts weekly.
Source · $100M Branding Playbook
Marketing Systems

SEO / Organic Growth

Gets StudyFix ranking for "how to revise cardiology finals" and similar queries.

Day-to-dayKeyword research, publishes disease-index micro-sites, builds backlinks, tracks GSC rankings, structured data.
Thrives if youlike slow-compounding wins, enjoy research, are patient with 3-6 month payoff windows.
Drains you if youneed instant feedback, hate technical setup (sitemaps, robots, schema), can't stand waiting.
StudyFix fitBig open runway. Per-module revision micro-sites plan to live at thestudyfix.com/<topic>/ — nobody owns them yet.
Source · $100M Marketing Machine (channel diversification)
Marketing People

Community Manager

Runs the WhatsApp cohorts, DM inbox, and eventually a Skool/Discord.

Day-to-dayAnswers every inbound DM, moderates cohort chats, spots which students are quietly disengaging, escalates the ones on fire.
Thrives if youlove people, remember names, feel energised by 1:1 messages, get satisfaction from someone saying "thank you".
Drains you if youfind message chats overwhelming, can't context-switch, hate being "on" all day.
StudyFix fitThe most immediate leverage seat. Every unread student DM = churn risk. Would be transformational if owned.
Source · $100M Marketing Machine (Communication & Communities scrape)
Sales People

Closer / 1-on-1 Sales Calls

Hops on 30-min calls with warm leads and closes them into premium tutoring.

Day-to-day3-6 discovery calls per day when full, pipeline management, follow-ups, objection handling.
Thrives if youlike the puzzle of a conversation, enjoy the moment of a signed £2k contract, don't mind rejection.
Drains you if youtake "no" personally, dread the phone, hate asking for money out loud.
StudyFix fitEarly-stage but real. Premium tutoring is 4-figures — every closed call moves the month. Youhana currently handles all of these.
Source · $100M Offers (framing) · $100M Fast Cash Playbook
Sales People

DM Setter / Front-of-House

Triage every inbound DM, book qualified ones onto Youhana's calendar.

Day-to-dayReads every inbound. Qualifies with 3 short questions, hands warm ones to the closer, thanks-but-not-nows the rest.
Thrives if youhave quick pattern-matching, love inbox zero, are naturally warm-but-firm.
Drains you if youare a slow decider, dread repetitive scripting, hate saying "no thanks".
StudyFix fitMassively underserved right now. Every reply that takes >12h is a lost lead.
Source · $100M Lead Nurture Playbook
Sales Systems

Retention & Renewals

Save calls. Win-back sequences. Off-boarding audits.

Day-to-dayWatches for churn signals, calls or DMs at-risk students, runs offboarding surveys, closes reactivations.
Thrives if youhave empathy without pushover-ness, are systematic, love the quiet-win metric of "saves".
Drains you if youfind leaving customers awkward, avoid tough conversations, don't like tracking cohorts.
StudyFix fitNot there yet — we need enough customers to lose first. Real seat by 2027.
Source · $100M Retention Playbook · $100M Lifetime Value Playbook
Product Systems

Head of Curriculum / Flashcard Design

Turns lectures into StudyFix decks & MCQs. The core StudyFix IP.

Day-to-dayUses Claude.ai + master prompts to draft cards, audits every card, exports as TSV, pushes to Anki.
Thrives if youare already a med student, care about the quality of your own learning, get a kick from a beautifully atomic card.
Drains you if youfind repetition boring, hate proofreading, don't like the feeling of "this could be 3% better".
StudyFix fitHighest overlap with what you already do for yourself. See Section 4 for the full deep-dive.
Source · StudyFix master prompt library · Anki v4 skeleton
Product Creative

Head of Media Production

The doing-hands. Editor, videographer, thumbnail designer.

Day-to-dayLives in CapCut / Final Cut for iPad. Turns raw footage into published clips across three ratios.
Thrives if youlove craft, enjoy tight loops of "cut / preview / re-cut", find editing meditative.
Drains you if youhate the finicky details, get RSI from tap-drag work, don't want to be a hidden hand behind someone else's face.
StudyFix fitOverlaps with Head of Content but is the pure production seat. Section 3 is the deep-dive.
Source · Section 3 of this playbook
Product Systems

Software / App Product Manager

Owns the flashcard app + SBA/OSCE app + lead magnet roadmap.

Day-to-dayUser feedback → prioritised feature list, works with Youhana on design pivots, runs QA before release.
Thrives if youare precise, enjoy "what should we build next?" more than "how do we build it?", love a great UX.
Drains you if youcan't tolerate ambiguity, want engineering to be someone else's problem entirely, hate reviewing screenshots.
StudyFix fitLead magnet is architecturally shipped but waiting on design pivot. This role unblocks the ship.
Source · StudyFix roadmap (pre-finals / post-exam phases)
Product People

Student Success Manager

Makes sure students actually pass. Weekly check-ins, results tracking.

Day-to-dayRuns 15-min weekly student calls, tracks Anki retention, spots the ones about to fail, escalates to Youhana.
Thrives if youget real energy from someone else's exam win, love the coaching side, are a natural mentor.
Drains you if youcan't emotionally separate from student outcomes, dread the tough "you're falling behind" call.
StudyFix fitDirectly ties to LTV. If a cohort passes finals, they pay again next year & refer three friends.
Source · $100M Lifetime Value Playbook
People Systems

Chief of Staff / EA to the CEO

Youhana's second brain. Meeting notes, follow-through, calendar, agenda enforcement.

Day-to-daySits in on every strategy call, writes decisions down, chases action items until they die, protects deep-work blocks.
Thrives if youlove making things run on time, get satisfaction from "closed the loop", enjoy proximity to strategy without owning it.
Drains you if youwant to be the face, can't nag politely, feel invisible without public credit.
StudyFix fitHighest-leverage seat inside the company. Leila explicitly named this as her earliest most-valuable hire.
Source · Leila's 62-page unfiltered advice · Q4 & Q26
People People

Hiring & Culture Ops

Recruits, onboards, and holds the culture bar as we hire our first 3-5 people.

Day-to-dayWrites job ads, screens candidates, runs interview loops, onboards new hires, updates the culture doc.
Thrives if youare a great judge of character, love asking hard questions, are willing to say "no hire" to someone Youhana likes.
Drains you if youfind recruiting logistics tedious, can't push back on the founder, dread awkward rejection emails.
StudyFix fitNot yet — we're 2 people. But the second we hire #3, this seat is created & whoever owns it defines the company for years.
Source · Leila's advice Q18, Q26, Q27 · $100M culture principles
Profit Systems

Bookkeeping / Ops Finance

Keeps business revenue & personal savings walled off. Reconciles Stripe ↔ Wise weekly.

Day-to-daySunday reconciliation, monthly export to accountant, tax jar sweeps, expense tagging, catches weird charges.
Thrives if youfind spreadsheets calming, love clean rows, get satisfaction from balanced books.
Drains you if youfind numbers stressful, hate detail work, resent "invisible" tasks nobody thanks you for.
StudyFix fitSmall time commitment (2-3h/wk) but non-negotiable. Currently ad-hoc & a liability.
Source · Section 5 of this playbook
Profit Analytical

Pricing & Offer Design

Runs the price of every StudyFix product. Tests, raises, packages, bonuses.

Day-to-dayBuilds offer stacks, tests price points, designs upsells, benchmarks against market rates for premium tutoring.
Thrives if youhave strong opinions on money, aren't afraid to charge more, love the strategic bit of "why this price".
Drains you if youfeel guilty asking for money, avoid discomfort, freeze on "what if nobody buys".
StudyFix fitDeferred until finals ship, but critical H2 2026. Pricing has been underpriced by 30-40% based on benchmarks.
Source · $100M Pricing Playbook · $100M Money Models · $100M Offers
Profit Analytical

Revenue Ops / Business Analyst

Builds the dashboards that tell you if you had a good month.

Day-to-dayCAC, LTV, cohort retention curves, monthly recurring revenue tracker, forecasting.
Thrives if youare naturally curious, love a good pivot table, enjoy the "what does this number really mean?" digging.
Drains you if youfind spreadsheets alienating, hate reporting nobody reads, want to be the one making decisions.
StudyFix fitNot yet — we don't have enough data. 2027 role, but worth knowing exists.
Source · $100M Money Models (unit economics chapter)
L

Leila's actual advice on picking your first role

"Where you are not strong, hire the experience. Where you are strong, bring people in that you can train to be your successors. The reality is you want to mix the two, and it is truly dependent on your and your co-founders' skillsets — what do you bring to the organization?"

— Leila Hormozi, 62-Pages of Unfiltered Advice, Q26
i

How to actually use this menu

  • Add as many as feel interesting — the shortlist can hold everything you're curious about, not just your top pick.
  • By end of week, narrow to 3. Hit Focus mode to hide the rest and compare only your picks side-by-side.
  • For each of the final 3, spend one afternoon doing it for real — a week of DMs, an afternoon of card design, a Sunday of bookkeeping.
  • Then we talk. The seat you actually want will be obvious after 3 real trials, not 3 hours of reading.
02 /

Curriculum · Prompt engineering & deck editing

Two sides of one skill — the prompt is the language going in, the deck is the language coming out. Path: "I've read some AI stuff on TikTok" → "I can audit a StudyFix prompt, find the language bug that's costing 8% coverage, run /verify-deck, and ship a patched deck to a real student." This is the #1 quality bottleneck in the business.

W1 Read what exists W2 Language audit W3 Small prompt fixes W4 First deck audits W5 Build a mini-prompt W6 Ship a real deck
WEEK1 Read the accumulated wisdom (no writing yet) ~4 hours · reading only
Read
  • 2_prompt_engineering/README.md — the tour
  • 2_prompt_engineering/learning-archive/LESSONS_LEARNED.md — the master doc; read every "What Failed and Why"
  • The feedback_* memory files in ~/.claude/…/memory/ — the curriculum in reverse
  • DO NOT read prompts/active/ or first-aid-app/src/lib/prompts/ — those are frozen. Read prompts/archive/ instead.
Do
  • Open 3 archived prompt versions (e.g. prompt2_v3.14.md, prompt2_v4.0.md) — read them next to each other
  • Write a paragraph on why v4.0 was rejected — LESSONS_LEARNED spells it out. Verify you understand each of the 6 reasons.
  • Screenshot two rules you don't understand yet — bring them to Youhana
Concepts to lock in
  • Echo test — cover the cloze, can you guess?
  • CHUNK vs SPLIT — one cloze per list vs distinct facts
  • Use-lecturer-language rule
  • 5-check cloze invariants (balanced brackets, no nested, no strays, no resets, no zero-cloze)
Ship: One page of notes titled "What I now believe about StudyFix prompts." Voice-note it if easier. Send to Youhana.
WEEK2 Language-issue audit — find bugs, don't fix them yet ~5 hours
Read
  • Pick one archived prompt (start with PROMPT_2_Anki_Creator_v2.2.md)
  • Reread the "Anti-Patterns" section of LESSONS_LEARNED.md
  • Memory: feedback_stem_language, feedback_use_lecture_language, feedback_anti_ai_slop_cards
Do
  • Copy the prompt into a Google Doc. Highlight in yellow every hedged verb ("might", "may", "consider", "aim to")
  • Highlight in pink every rule that tells the model what to think instead of what to output
  • Highlight in blue every place that mixes two decisions in one sentence
  • Highlight in green every place where the prompt already does something great — steal the pattern
Look for
  • Rules that echo-test-fail: does the stem tell you the answer?
  • Rules that mandate cards for lectures that don't have that content type
  • Verification steps embedded in the prompt (belong in Python)
  • Word limits on clozes ("~5–10 words") — anti-pattern per LL
Ship: One annotated Google Doc with 10+ highlights. No fixes yet. Just diagnosis.
WEEK3 Small fixes on archived prompts (safe to break) ~5 hours
Read
  • Memory: feedback_if_then_cloze_structure, feedback_cloze_scope_expansion, feedback_no_master_overview_duplicates
  • Reference: pipelines/templates/FORMATTING_REFERENCE_v1_2026-06-14.md — Recipe A + 3-pattern taxonomy
Do
  • Take 3 of your Week 2 highlights. Rewrite each rule to the memory-file standard.
  • For each rewrite: state what you changed and why in one line. If you can't, revert.
  • Test one rewritten prompt on a non-critical lecture (an old ICC-1 archive PDF). Do not use a fresh lecture — that's live product.
  • Print 5 cards from the output. Do the echo test on each. Which pass, which fail?
Failure modes to catch yourself doing
  • Paraphrasing rules — v4/v4.1 died from this. Preserve verbatim when possible.
  • Adding process to fix quality — every added step costs model attention
  • Fixing a symptom, not the root — if 5 cards fail the echo test, one rule is broken, not 5
Ship: A short before/after doc — 3 rules, the fix, the reasoning. Show Youhana; get red-pen feedback.
WEEK4 First deck audits — the output side of the same skill ~6 hours
Read
  • Full verify-deck skill spec (in .claude/skills/)
  • eval_cards.md — the 12-category rubric
  • audit_cards_completeness.md + audit_cards_synthesis.md
  • Memory: feedback_recipe_a_visual_format, feedback_no_master_overview_duplicates
Do
  • Print the 12 categories on one sheet. Tape to your desk.
  • Pick 3 archived decks from 3_flashcard_app/archive/first-batched-cards/
  • Run /verify-deck on each. Note patches · drops · rewrites.
  • Cross-check the AI's audit against your Week 2 language-issue instincts. Where do you disagree?
  • Learn lecture-mode detection (LGW / PATH / STANDARD) — classify 5 lecture PDFs, justify each in one sentence
Watch for
  • False patches — "add this card" for content that's already in the deck under another name
  • Missed images — X-ray / CT / ECG / histology slides need a photo-describe card
  • List fragmentation — a "name all 5" answer split into 5 micro-clozes = wrong (memory: feedback_list_as_unit_one_cloze)
  • Mode mismatch — PDF signal = LGW but deck = topical → verify-deck must stop and ask. The ICC-1 [039] incident happened because nobody caught this.
Ship: Three deck-audit summaries in a shared Notion. Format: missing content · redundant cards · rewrite candidates · format issues · one surprise. Youhana marks his diff.
WEEK5 Build a small prompt from scratch (non-production) ~6 hours
Read
  • The blurt-cues skill spec — it's a good small prompt to reverse-engineer
  • Memory: feedback_cloze_strict_invariants, feedback_split_distinct_facts
Do
  • Pick a small self-contained job: "turn a lecture PDF into a 5-question OSCE mini-viva"
  • Draft the prompt. Include: role, scope, output format, 3 examples, 3 anti-examples, output schema.
  • Run it on 3 different lectures. Same output every time? Log what drifts.
  • Add rules only where drift shows up — not preemptively
Success looks like
  • Prompt is under 800 words
  • Every rule can be defended with "if I remove this, X breaks"
  • Output is copy-pasteable into an OSCE study session with zero editing
Ship: A working mini-prompt saved to 2_prompt_engineering/prompts/candidates/manom-osce-viva-v0.1.md. Youhana runs it. You watch him use it.
WEEK6 Ship a real deck end-to-end + pair on a live audit ~5 hours + 1 real session
Read
  • Memory: feedback_cloze_strict_invariants — the 5-check pre-ship audit
  • Memory: feedback_cloze_fill_no_mechanical_shortcuts, feedback_reset_context_between_lectures
  • The Surgical_Triage_playbook reference
Do
  • Sit with Youhana while he runs verify-deck on one real lecture. Take notes.
  • Next lecture: you drive the audit. He watches.
  • Flag patches. Debate one drop. Argue one rewrite.
  • Apply additive patches. Drop redundant. Rewrite flagged.
  • Run the 5-check invariant audit: balanced brackets · no nested opens · no stray closes · no number resets · no zero-cloze.
  • Ship the deck to Desktop\ICC-2\super_cloze\<module>\. Ship a separate delete-list — Anki never auto-removes.
Graduation criterion
  • You look at a card and name which memory rule it violates in under 30s
  • You defend a prompt rule against "why is this even here?"
  • You rewrite a bad stem into a good stem without asking
  • You catch a bleed-through hallucination that AI-only would have shipped
Ship: One real deck imported to a student's Anki with your name on it. Youhana signs off. You now own prompt-and-deck QA for one full module.

The AI workflow for prompt engineering

Primary tool: Claude Pro (Max plan) — StudyFix runs on Claude, so learn the tool that ships product. Secondary: Claude Code CLI for editing the actual prompt .md files.

  1. Set up a Claude Project called "SF Prompt Lab". Drop LESSONS_LEARNED.md, 3 archived prompts (never the active ones), and your top 5 feedback_* memory files as project files.
  2. For every rewrite you draft: paste it in, ask Claude to critique against LESSONS_LEARNED. It'll flag paraphrasing anti-patterns and verification-in-prompt smells fast.
  3. Use the Anthropic Console workbench for playground testing — run two prompt variants on the same lecture side-by-side, compare outputs cell-by-cell.
  4. For real deck audits, use the /verify-deck skill inside Claude Code. That IS the audit workflow — no need to rebuild it.
  5. Never paste live student PDFs into a public Claude chat. Use archived material only until you own Claude Max with the right data policy.

Best-value resources

Ranked by signal-per-hour, not popularity
Anthropic Courses Free Start here. Anthropic's own prompt-engineering + Claude Code courses. Right tool, right teacher, right examples. DeepLearning.AI · Short Courses Free Andrew Ng's platform. 1–2 hour bites. "Prompt Engineering for Developers" + Claude-specific courses. Simon Willison's blog Free The single best source on real-world LLM engineering. Read the archive weekly, not once. Latent Space Free Podcast + newsletter. What founders are actually shipping this week. Skip the hype; keep the technique.
◆

Graduation — what "done" looks like at the end of Week 6

Both sides of the language layer are yours. You read the archive, mapped bugs, fixed prompts, audited decks, built a mini-prompt from scratch, and shipped a real patched deck. Next: you own prompt-and-deck QA for one whole subject (e.g. cardio finals). You don't write the pipeline — you keep it honest.

03 /

Curriculum · Video editing on iPad

Not "become a filmmaker" — become fast enough to turn a batch-film day into 5 clips a week that don't feel like AI slop. CapCut on iPad is the primary tool. Every clip ships in three ratios and passes a medical-term caption audit.

W1 CapCut foundation W2 Hooks & pacing W3 Overlays & captions W4 The three ratios W5 Ship a whole week
WEEK1 CapCut foundation — cuts, silence, pacing ~4 hours
Install & set up
  • CapCut for iPad (free) — the primary tool
  • iCloud shared album with Youhana: SF-Raw
  • Folder discipline: /Files/StudyFix/Raw/YYYY-MM-DD/ → /Ready/
Do
  • Take one clip from an old batch-film day. Cut it end-to-end.
  • Rules: silence & ums cut aggressively, jump-cut every 2–3 seconds, first 1.2s earns the scroll
  • Target duration: 22–34s for Reels/TikTok; Shorts <60s hard cap
  • Compare your cut to the original — how much did you remove? (Target: 40%+)
Accounts to study (60 min)
  • @personalbrandlaunch — the archetype for a solo-founder brand run right. Watch 5 top Reels back to back and screenshot the openings.
  • Ali Abdaal — adjacent-aesthetic, med-student roots
  • Peter McKinnon "how to edit fast" videos — technique
  • Ask Youhana for his top 3 study accounts of the month — the shortlist rotates
Ship: One 22–34s clip, edited. Youhana watches it back once, gives one note.
WEEK2 Hooks & pacing — the first 1.2 seconds ~5 hours
Read
  • 1_marketing/01_marketing/Hook_Database.md
  • 1_marketing/01_marketing/AD_HOOKS_BANK.md
  • Memory: feedback_brand_voice_anti_ai_slop
Do
  • Take one raw clip. Cut 5 different hooks from the same source (first 1.2s each).
  • Cover test: with the rest of clip hidden, does the hook alone earn the scroll?
  • Try each hook type: claim, question, freeze-frame, contradiction, before/after
Voice rules
  • No "hey guys." No "in this video I'll..."
  • Founder-honest — anchor: "Most study advice is bullshit"
  • No "AI-powered" / "seamless" / "unlock" / "empower" / "✨"
Ship: Same clip in 5 hook variants. A/B panel — Youhana picks his top 2.
WEEK3 Overlays, captions & medical-term audit ~5 hours
Do
  • Screenshot 5 real Anki cards from an archived deck. Add each as an overlay synced to the moment the claim is made in the audio.
  • On-screen text: SF Pro Rounded, 90–120pt, white on 40% black scrim. No emojis in on-screen text.
  • Enable CapCut's "Reading Time" progress bar template.
Captions — the medical audit
  • Run CapCut auto-captions in English (UK).
  • Scrub every drug name, anatomy term, eponym — Wernicke, Cushing, Kussmaul, warfarin dose, potassium value.
  • Break lines at phrase boundaries — never mid-word, never mid-drug-name.
  • Text stays in the middle 60% vertical band so the IG UI doesn't cover it.
Non-negotiable
  • Zero tolerance on medical typos in captions. Screenshotted mistakes go viral for the wrong reason.
  • Assume every published clip is evidence in a hearing.
Ship: One clip with Anki-card overlays + fully audited captions. Ready to post.
WEEK4 The three ratios — export & schedule ~4 hours
Do
  • Take your Week 3 clip. Export in three ratios:
  • • 9:16 · 1080×1920 — Reels, TikTok, Shorts
  • • 1:1 · 1080×1080 — LinkedIn, Twitter inline
  • • 16:9 · 1920×1080 — YouTube long only (monthly)
  • Reformat on-screen text safe zones per ratio
  • Naming: SF-042_hook_9x16_v1.mp4
Schedule
  • Later — IG + TikTok + LinkedIn
  • Metricool — Twitter/X + Threads
  • Reels fire at 12:45 UK, TikTok +15 min later
  • LinkedIn carousel at 13:30 UK
  • Pre-fill the first pinned comment — CTAs live there, not captions
Log everything
  • URL per platform in Notion Content Tracker
  • Hook (first 6 words), format tag, hashtag set
  • First-hour views screenshotted into the row
Ship: One clip live across all channels + first-hour metrics logged.
WEEK5 Own a full posting week ~10 hours across the week
Do
  • Take a full batch-film day (30+ clips). Edit 5.
  • Schedule across all channels — Later + Metricool both show a green 7-day queue
  • Log every post in the Notion Content Tracker
  • Friday: retro on which 5 performed best. Save top 3 to inspiration folder.
Graduation criterion
  • Youhana doesn't touch the calendar. It runs itself.
  • You can turn one raw clip into 3 ratios in under 45 minutes
  • Zero medical typos have shipped that month
  • You've replicated one hook pattern from a Casey / Colin&Samir teardown
Watch for burnout
  • Editing is deceptively tiring — batch on 2 fixed days, not "a bit every day"
  • Kill perfectionism — ship at 85% quality daily beats ship at 95% weekly
Ship: A full week's queue turned green. Friday retro sent to Youhana with the 3 best clips highlighted.

The AI workflow for video editing & content

Primary tools: Claude Pro (hook drafts + caption audit), ChatGPT Pro (thumbnails via DALL-E + voice-mode brainstorm). Optional: Descript ($12/mo) for AI-based filler removal on iPad.

  1. After each batch-film, dump the raw transcript into a Claude Project called "SF Hooks Lab". Load Hook_Database.md, AD_HOOKS_BANK.md, and the anti-AI-slop memory as project files so voice is baked in.
  2. Ask Claude: "Give me 10 hook variants for this clip in the voice defined by the project files." Screenshot the ones that pass the cover-test.
  3. Run CapCut auto-captions → paste into Claude → "Flag every medical term that looks misspelled or ambiguous." This is your zero-tolerance safety net for drug/eponym typos.
  4. For thumbnails, use ChatGPT Pro + DALL-E: describe the concept in one sentence, generate 4, pick 1, refine.
  5. For rubber-ducking a stuck edit, open ChatGPT voice mode and talk through the problem for 5 minutes. Faster than typing.
  6. Never let AI write the final on-screen text. Draft-yes; ship-no. Every word overlaid gets your eyes on it.

Editing craft — best-value

iPad-first, short-form-first, free-first
CapCut official tutorials Free Start here. The tool StudyFix uses daily. 90 min tutorial gets you 80% of the way. Peter McKinnon Free Editing speed + cinematic sense. His "edit faster" videos alone save you 5 hours a week. Casey Neistat's original vlogs Free Pacing masterclass by osmosis. 3 old vlogs at 1.5× and you internalise cut rhythm. Justin Odisho Free Practical editing tutorials. If you outgrow CapCut, DaVinci Resolve on iPad starts here.

Personal branding & content strategy — the titans

Learn the voice, the framing, the business model behind the videos
Caleb Branston Free The archetype. Free YT courses on personal branding + creator business. Watch his "how to build a personal brand" playlist top-to-bottom. Alex Hormozi Free The StudyFix brand voice anchor. Free content that's better than most $1000 courses. Focus on the "$100M Leads" playlist. Colin & Samir Free Creator economy insiders. "How top creators edit" series is 3 episodes of pure signal. Also great podcast interviews with the biggest YouTubers. Dan Koe Free Essayist-creator hybrid. The strategic thinking behind why a creator becomes a business. Best watched at 1.5×. Justin Welsh Free The solo-creator playbook. LinkedIn + newsletter + digital products. Directly relevant to StudyFix's founder-brand approach. Chris Do · The Futur Free Brand + design fundamentals. His "how to charge more" and "brand strategy" playlists are what every founder should watch once. Ali Abdaal · Part-Time YouTuber Academy Premium option. Only if StudyFix invests — then it's the most creator-tailored course you'll find. Med-student roots = adjacent brand.
◆

Graduation

You own the entire distribution surface for StudyFix's short-form content. Youhana films, you edit, the calendar runs green, and no medical typo has shipped this month. The 12:45 UK Reel drops without a Slack ping.

04 /

Curriculum · Vibe coding + App PM

Not "become a software engineer." The goal: ship product changes without needing Youhana in the seat. You'll own the flashcard app + SBA/OSCE app + lead-magnet roadmap. GitHub Codespaces + Claude/Copilot writes the code — you drive. Fully iPad-doable in Safari.

W1 Set up the seat W2 First tiny PR W3 First real feature W4 Ship a design-pivot chunk W5 Own the release cadence
WEEK1 Set up the seat ~4 hours
Set up (iPad-only)
  • GitHub app on iPad (view PRs, issues) + Working Copy for local git ($20 pro).
  • Get invited to the StudyFix GitHub org by Youhana.
  • Open the repo in GitHub Codespaces in Safari — full VS Code in-browser. Codespaces auto-installs Node + pnpm.
  • Run pnpm install && pnpm dev in the Codespace terminal. App previews at a private Codespaces URL — open it in another Safari tab.
Read
  • 4_sba_osce_app/studyfix-lead-magnet/README.md
  • docs/ROADMAP.md — what's next
  • docs/HANDOFF_design_pivot_amboss_uworld_2026-06-19.md — the active pivot
  • docs/DEPLOY_CHECKLIST.md
Concepts to lock in
  • Feature branches, never main — you always branch, always PR
  • Vercel preview URLs — every branch auto-deploys a preview link
  • 92/92 tests must stay green — the pre-finals invariant
Ship: Screenshot of the app running via the Codespaces preview URL in Safari. Text Youhana "seat's warm."
WEEK2 First tiny PR — copy change, no logic ~3 hours
Do
  • Find one string on the landing page that sounds AI-slop (e.g. "seamless", "unlock", "empower")
  • Ask Copilot Chat inside the Codespace (or Claude.ai in the other tab) to change it to on-brand voice (memory: feedback_brand_voice_anti_ai_slop)
  • Read the diff line by line. Only accept if you understand every changed line.
  • Push to branch. Open PR. Merge after tests pass.
The AI-conversation to have
  • "Change the H1 on the landing page from X to Y. Do not touch anything else."
  • Constrain the model — narrower prompt = safer diff
  • If the diff includes files you didn't ask about, reject the PR and reprompt
Success looks like
  • PR is under 5 lines changed
  • You can defend each line to Youhana in one sentence
  • Tests are still 92/92
Ship: One merged PR. Vercel preview link in Youhana's DMs. Congrats — you shipped code.
WEEK3 First real feature — a component behind a flag ~6 hours
Read
  • Existing components in studyfix-lead-magnet/src/components/ — pick 3, read till you understand
  • How feature flags currently work in the codebase (grep for FLAG_ or check docs/)
Do
  • Pick one small user-facing feature from ROADMAP.md
  • Draft a 6-line spec first: who, what, screen, edge cases, success criterion, kill criterion
  • Prompt Copilot Chat (or Claude.ai) with the spec, not the implementation
  • Test on iPad Safari before merging (memory: browser test isn't optional)
Rules of engagement with the AI
  • Never accept "here's a fix" without reading the diff
  • If you don't understand a line, ask the AI to explain it before merging
  • If the AI writes a comment, delete it unless the WHY is non-obvious (memory: writing-comments)
Ship: The feature behind a flag, off by default. Youhana turns it on if it looks right.
WEEK4 Ship a chunk of the AMBOSS/UWorld design pivot ~8 hours across the week
Read
  • docs/HANDOFF_design_pivot_amboss_uworld_2026-06-19.md — top to bottom
  • Memory: feedback_design_split_landing_workspace — landing editorial, workspace clinical
  • Screenshot 5 UWorld + 5 AMBOSS screens as visual reference
Do
  • Pick one screen from the pivot backlog — the smallest one
  • Draft a before/after Figma or Excalidraw sketch on iPad
  • Build it. Ship it. Deploy preview link.
  • Sit with Youhana while he uses it. Note every pause, every squint, every "hmm."
The rule
  • Landing = editorial (bold, opinionated, magazine-like)
  • Workspace = clinical (dense, quiet, exam-tool)
  • Never unify them. Don't relitigate. Ship on-brand for the surface.
Ship: One pivoted screen live on the preview URL. Screenshot in the roadmap doc.
WEEK5 Own the weekly release cadence ~1 hour/day
Do
  • Every Monday: pick 3 issues from the roadmap. Post them in Youhana's DMs as your week.
  • Every Friday: ship a release note. Format: what shipped · what didn't · what broke.
  • Every failing test: your problem, not Youhana's. Debug or roll back within 2 hours.
Own the roadmap
  • Read docs/ROADMAP.md weekly — reorder items
  • Kill items that no longer make sense — half-finished isn't better than nothing (memory: no-half-finished)
  • Say no when Youhana overloads the sprint
Graduation criterion
  • Youhana goes quiet for 3 days and the app keeps shipping
  • You call for the design/product/tech decision, not him
Ship: A 4-week release-note log. If it reads like a real product's changelog, you graduated.

The AI workflow for vibe coding + PM (iPad-first)

Primary stack (iPad-only, zero desktop): Claude.ai Projects in Safari + GitHub Codespaces in Safari (full VS Code in-browser) + v0.dev for UI mockups + ChatGPT voice-mode for rubber-ducking. Escalation only: Blink Shell → SSH into a Mac/cloud VM for real Claude Code CLI. Not required to graduate.

  1. Set up a Claude Project called "SF App PM". Drop ROADMAP.md, README.md, DEPLOY_CHECKLIST.md, and HANDOFF_design_pivot_amboss_uworld_2026-06-19.md as project files.
  2. For every feature: write a 6-line PM spec first in the project (who · what · screen · edge cases · success · kill criterion). Only then open a Codespace.
  3. Spin up a Codespace on the studyfix-lead-magnet repo — Safari opens VS Code. pnpm install && pnpm dev runs; the dev server previews at a private Codespaces URL.
  4. For AI edits inside the Codespace: use GitHub Copilot Chat (built-in) for inline diffs, or paste code back to the Claude Project tab for tricky refactors. Read every line before accepting. Unrelated files touched → reject and reprompt.
  5. UI mockups: v0.dev in a third Safari tab. Describe the component in prose, get a React skeleton, paste into the Codespace.
  6. Stuck in a loop? ChatGPT voice-mode on iPad — talk it out for 5 min. Reframes better than a fresh chat.
  7. Never let the AI touch prompts/active/ — CLAUDE.md off-limits rule. Kills prompt regressions.

Best-value resources

Two tracks: (a) drive the AI well · (b) product-thinking
Anthropic Claude Code course Free Start here. The right teacher for the tool you'll use every day. Watch end-to-end, then rewatch weeks 1–2. GitHub Codespaces Docs Free iPad-primary tool. Full VS Code in Safari, pnpm/node preinstalled, dev server preview URLs. Read the "getting started" + "port forwarding" pages first. GitHub Copilot Chat Your AI-in-editor inside the Codespace. Free for verified students via GitHub Student Pack — check your KCL email first. Joy of React (Josh Comeau) Best-in-class React course. Deep understanding of what the AI is writing for you. Worth it if you're serious about the lane. freeCodeCamp Free The cheap alternative to Joy of React. React + TypeScript full-length tutorials. Not as polished — same fundamentals. Refactoring UI Adam Wathan + Steve Schoger. How to make things look good without a designer. Pays back on the very first landing page. Inspired (Marty Cagan) The product-management bible. Read chapters 1–5 in one sitting, skim the rest. Better than any $2k PM bootcamp.
◆

Graduation

You own the app roadmap. You can ship a small feature end-to-end without Youhana in the loop. The AI writes the code — you make the calls. StudyFix now has an actual PM.

05 /

Curriculum · Bookkeeping + revenue ops

The pillar that keeps the business alive. You'll learn to reconcile Stripe against Wise, keep the business/personal wall bulletproof, build the first MRR dashboard, and compute the numbers Youhana currently guesses (CAC, LTV, cohort retention).

W1 Access + the wall W2 Sunday reconciliations W3 First MRR dashboard W4 CAC / LTV W5 Monthly close
WEEK1 Access + the business/personal wall ~3 hours
Get
  • Stripe read-only access — Youhana adds you as "Analyst"
  • Wise Business viewer access
  • Notion page: "SF Ledger" — you own this tab
Read
  • 1_marketing/02_business/FULL_BUSINESS_AUDIT_2026-05-31.md — the baseline
  • 1_marketing/02_business/MONEY_MODEL_DESIGN.md
  • FOUNDING_TUTOR_OFFER.md — the pricing anchor
The wall
  • Business revenue → Wise Business only. Never touches personal savings.
  • Any exception → logged in SF Ledger same day, with reason
  • 25% of every payout auto-swept to SF-Tax jar in Wise
  • All expenses on the business card, tagged in Wise
Ship: Screenshots of both dashboards + a Notion Ledger with the wall rules pinned. You now know where every pound lives.
WEEK2 Four Sunday reconciliations in a row ~1 hour/wk × 4
Do (Sun 18:00)
  • Open Stripe. Note payouts settled that week.
  • Open Wise. Match each payout line.
  • Any mismatch → tag, screenshot, message Youhana same-day.
  • Log to Notion Ledger. Screenshot the reconciled state.
Learn
  • Stripe fees ≠ payout amount. Learn the fee structure (2.9% + $0.30 UK equivalent).
  • Refunds show as negative payouts. Failed charges show as retries.
  • Chargeback vs refund vs dispute — three different things.
Categories
  • Founding Tutor Pack ($500)
  • Continuity ($40/mo)
  • Concierge Day (separate line)
  • Refunds / chargebacks (auto-flagged)
Ship: Four weeks of reconciled Sundays in the Notion Ledger. Zero mysteries.
WEEK3 Build the first MRR dashboard ~5 hours
Read
  • 1_marketing/02_business/REVENUE_FORECAST_30_60_90.md
  • 1_marketing/02_business/PRICING_TEST_EXECUTION_PLAN.md
Do
  • Google Sheet or Notion database. Rows = weeks. Columns:
  • • New tutors (Founding Pack sales)
  • • Continuity MRR (active $40/mo)
  • • Concierge Day revenue (separate line)
  • • Total MRR
  • • Churn (canceled continuity)
  • • Net new MRR (new + reactivations − churn)
  • Populate 8 weeks back from Stripe. That's your baseline.
Rule
  • One-time sales are NOT MRR. The $500 pack is amortised over its lifetime; only the continuity $40 counts as recurring.
Ship: A live dashboard link. Every Sunday, the top row updates. Youhana bookmarks it.
WEEK4 Compute CAC / LTV / cohort retention ~6 hours
Read
  • 1_marketing/02_business/MARKET_BENCHMARKS_2026-05-31.md
  • 1_marketing/02_business/REVENUE_FORECAST_30_60_90.md
Do
  • CAC — total spend in a month ÷ new tutors that month. Include content time at a shadow-rate if you're being honest.
  • LTV — average tutor's total lifetime spend ($500 pack + N months at $40)
  • Cohort retention — of tutors who bought in Month N, how many still paying continuity in Month N+1, N+2, N+3?
  • Compute the LTV:CAC ratio. If < 3, we have a problem. Say so.
Honest failure modes
  • Not enough tutors yet for statistical significance — say so, don't fake it
  • Founding pack lifetime lock inflates future LTV — model both scenarios
  • Content time is real cost — count it even though it feels free
Ship: A 1-page "State of the Numbers" doc. Best guess, biggest unknown, biggest lever. Send to Youhana.
WEEK5 Own the monthly close ~4 hours end of month
Do (last Sunday of month)
  • Reconcile the full month Stripe ↔ Wise
  • Export expense report from Wise → send to accountant
  • Confirm SF-Tax jar was auto-swept correctly
  • Write a 1-page P&L: revenue lines, cost lines, net
  • Update MRR dashboard, CAC, LTV, cohort curves
  • Send the whole packet to Youhana with a 3-bullet "what changed vs last month"
Graduation criterion
  • Youhana reads your monthly close and knows more than he did before
  • You can answer "how are we doing?" in three sentences without hedging
  • You spot the trend before he does
What you now own
  • The wall (business/personal)
  • The reconciliation ritual (weekly)
  • The dashboard (live)
  • The monthly close (packet to Youhana + accountant)
  • The instinct for whether the month was good
Ship: The first monthly close packet. If it takes > 4 hours, the process is broken — fix the process next month.

The AI workflow for books & rev ops

Primary: Claude Pro (memos, models, one-page P&Ls). Research: Gemini Pro Deep Research (benchmarks, market comparables). Spreadsheets: ChatGPT Pro Code Interpreter (CSV pivots, one-off analyses).

  1. Set up a Claude Project called "SF Finance". Load monthly Stripe + Wise CSV exports as project files. That's your baseline for every question.
  2. Ask Claude: "Compute MRR, new tutors, churn, and total revenue for this month vs last. Give me a 5-line summary and one insight I would have missed."
  3. For research questions ("what's the CAC benchmark for premium tutoring in the UK?"), use Gemini's Deep Research mode — it'll auto-crawl 30+ sources and cite them.
  4. For quick pivot tables on a Stripe CSV, use ChatGPT Pro Code Interpreter — upload, describe the pivot in one line, download the result.
  5. Every AI-drafted number gets eyeball-verified against the raw Stripe dashboard before it lands in a doc Youhana reads.
  6. Never paste customer PII into a public chat. Anonymize (replace names/emails with codes) before uploading. If unsure, ask Youhana first.

Best-value resources

Bookkeeping fundamentals + SaaS metrics literacy. Skip the CFO bootcamps — waste.
Xero Central Learning Free Start here. Bookkeeping fundamentals + UK small-business SOPs. Even if you never touch Xero, the framework is right. David Sacks · The Cadence Free SaaS operating rhythm. One essay, read once a quarter. Shapes how you think about weekly / monthly / quarterly close. Paddle / ProfitWell blog Free SaaS metrics explained by people who eat them for a living. MRR, churn, LTV — no MBA fluff. ChartMogul Institute Free Metric deep-dives + dashboard templates. When you build the MRR dashboard in Week 3, this is your reference. The E-Myth Revisited (Gerber) How small businesses actually work vs how founders think they work. Read on holiday. Changes how you see StudyFix. Warren Buffett's shareholder letters Free Accounting intuition without an MBA. Read the last 5 letters. You'll spot financial nonsense in every pitch deck after.
◆

Graduation

Money moves through the business with a human watching it. Youhana stops guessing at MRR. StudyFix has an actual finance function — one person, four hours a month, tight numbers.