← Generative AI Mastermind InternationalAll programsHomeSearch
Generative AI Mastermind International·Mastermind Session Recordings·8:28:56

Day 2: Custom GPTs and Gems, Markdown Prompting, Cloning Your Writing Style, MCP in Daily Use, a Vibe-Coded Calorie Counter, and the Consultant-to-SaaS Road

Vaibhav Sisinty Mentor - custom GPTs, markdown prompting, the AI-mentor bot, MCP and the level 1-5 personal workflow (0:32-3:26) · Dileep Mentor - product architecture and the vibe-coding build (5:41-7:10); the monetisation roadmap (7:49-8:18) · Priyatam Vusala Host ('PV') - recap, sales pitch, Q&A, bonuses

The short version

  1. Custom GPTs by conversation: describe the bot, iterate on name, tone and a scoring table, then open Configure to see the generated system prompt (0:32-1:03). Markdown prompting - Role, Objective, Context, Instructions, Notes with # headings and *emphasis* - is the upgrade to an 'advanced GPT' (1:03-1:12).
  2. Clone your own voice: feed 9-10 of your posts, ask for a content-DNA report, have a prompt-engineer persona turn it into a markdown system prompt, paste into a GPT or a free Gemini Gem; 'gaslight' the model with a rival's 9.5/10 for a better second pass; ask for the prompt 'in a code block' so the markdown survives (1:13-1:41).
  3. An AI mentor from the Day-1 transcript (Happy Scribe, 335 pages) as a Gem's knowledge; the same agent rebuilt in Lyzer for an API, given a chat UI by Emergent in 20 minutes, and turned into a VAPI voice agent with a cloned voice (1:34-2:07).
  4. Level 1-5 as one person's practice: Bolt AI over OpenRouter and Ollama with visible model parameters; Apify MCP scraping an Instagram profile's reels into a LinkedIn post; Goose recommending disk clean-up without deleting; a VAPI MCP phone call placed live; a Flux LoRA face and voice clone running an Instagram to 1.2M followers; 'Jerry' the executive-assistant agent; a three-day vibe-coded CRM instead of a $120k HubSpot quote (2:08-3:26).
  5. Dileep's calorie-counter build: the movie analogy for architecture; Perplexity competitor screenshots -> feature table -> PRD -> trimmed build prompt -> Replit (Genspark in parallel) -> a wrong-vision-model bug fixed by rolling back to GPT-4o -> one feature at a time, checkpoints, Publish (5:41-7:10). Then the money: consultant -> agency -> service-as-software -> SaaS, a 10-point filter for profitable AI niches, and where to find dead ideas worth rebuilding (7:49-8:18).

At a glance, three clicks deep

Skim here first: the closed row is the glance, open is the study card with the key points and timestamps, and the ↓ link drops to that concept's full write-up below.

01Markdown prompting: Role, Objective, Context, Instructions, NotesHeadings + emphasis + five sections;›

Headings + emphasis + five sections; the difference between a generated GPT prompt and an advanced one.

# headings, * emphasis (1:04-1:06)

Role / Objective / Context / Instructions / Notes (1:06-1:12)

MIT-graduate 'why it matters' analogy (1:09)

↓ Full write-up of this concept

02Build a custom GPT by talking to the wizard, then read what it wroteConversational wizard -> preview -> refine -> Configure to see and edit the prompt.›

Conversational wizard -> preview -> refine -> Configure to see and edit the prompt.

Describe, name, logo, iterate (0:40-0:50)

Scoring + ranking table added in plain English (0:50-0:54)

Configure reveals the system prompt (0:55-0:58)

Gemini Gems as the free equivalent (1:29-1:34)

↓ Full write-up of this concept

03Cloning your writing style: content DNA -> prompt engineer -> GPT, with the gaslight and the code-block tricksDNA report -> prompt engineer -> saved bot;›

DNA report -> prompt engineer -> saved bot; gaslight for effort; code block for raw markdown.

9-10 posts -> content-DNA report (1:13-1:17)

Second chat converts it to a markdown prompt (1:18-1:23)

'Claude did 9.5/10' trick (1:19, 1:26)

Ask for a code block to keep the markdown (1:39-1:41)

↓ Full write-up of this concept

04From a transcript to a mentor bot, an app, and a phone voice: Gem -> Lyzer -> Emergent -> VAPIKnowledge-grounded Gem -> Lyzer agent with API -> Emergent UI -> VAPI voice.›

Knowledge-grounded Gem -> Lyzer agent with API -> Emergent UI -> VAPI voice.

Transcript as knowledge; escalation rule (1:36-1:44)

Lyzer agent -> Agent API (1:46-1:50)

Emergent chat UI in ~20 min (1:50-1:54)

VAPI assistant with cloned voice (1:55-2:07)

↓ Full write-up of this concept

05Levels 1-5 as daily practice: Bolt AI + OpenRouter + Ollama, Apify and VAPI over MCP, Goose, a clone, Jerry, a CRMModels and parameters -> MCP automations -> multimodal clones -> personal agents -> vibe-coded products.›

Models and parameters -> MCP automations -> multimodal clones -> personal agents -> vibe-coded products.

Bolt AI over OpenRouter and Ollama; parameters (2:12-2:20)

Apify MCP content pipeline (2:24-2:36)

Goose file clean-up; VAPI call live (2:32-2:41)

Flux LoRA clone; the Instagram numbers (2:42-2:50)

Jerry; the $120K-vs-3-day CRM (2:50-3:15)

↓ Full write-up of this concept

06Product architecture as a movie: front end, back end, API, database, integrations, authSix roles in one metaphor;›

Six roles in one metaphor; read any app's architecture with it.

Front / back / API / database (5:46-5:50)

Integrations as VFX; the CGI mismatch (5:50-5:52)

Auth as ticketing (5:53)

↓ Full write-up of this concept

07Vibe-coding a calorie counter: competitor screenshots -> PRD -> trimmed prompt -> Replit -> debug -> rollbackResearch -> PRD -> constrained prompt -> build -> factual debugging -> checkpoints -> publish.›

Research -> PRD -> constrained prompt -> build -> factual debugging -> checkpoints -> publish.

Competitor feature table from screenshots (5:56-6:05)

PRD then a trimmed guideline prompt (6:05-6:12)

Vision-model bug fixed by rolling back to GPT-4o (6:22-6:35)

One feature at a time; checkpoint rollback (6:50-7:08)

↓ Full write-up of this concept

08Consultant -> agency -> service-as-software -> SaaS, and the 10-point filter for profitable nichesFour-year path compressible to one or two;›

Four-year path compressible to one or two; ten filters for where AI pays.

Consultant -> agency -> service-as-software -> SaaS (7:52-7:58)

Ten-point niche filter (7:58-8:03)

30-day execution playbook (8:14-8:18)

↓ Full write-up of this concept

09Where to find dead ideas worth rebuilding with AIDemand proven, execution died;›

Demand proven, execution died; rebuild it cheaper with AI.

Product Hunt graveyard, abandoned extensions (8:09-8:11)

Indie Hackers plateaus, dead YC startups (8:11-8:13)

Stale GitHub repos - check the licence; Reddit complaints (8:13-8:14)

↓ Full write-up of this concept

The concepts in full

01

Markdown prompting: Role, Objective, Context, Instructions, Notes

Plain text has no bold. Headings and asterisks are how the model sees emphasis.

Use # / ## / ### to rank sections and *word* for emphasis, because the model reads formatting weight the way a human would. Five sections: Role (persona), Objective (the task), Context (why it matters - the MIT graduate who will not take a job seriously until told why), Instructions (numbered), Notes (everything else). Demoed on a customer-success agent for Outskill's inbox. It is the same framework BC9 Day 3 teaches for the same reason.

Why it matters

The structural prompt template behind every bot built on this day.

02

Build a custom GPT by talking to the wizard, then read what it wrote

A Viral Tweet Generator in ten minutes: name, logo, tone, then 'rate each 1-10 for virality, brutality, relatability and sort'.

PROCEDURE: Create -> describe the bot in natural language (dictated) -> accept or change the proposed name and logo -> test in the preview -> refine in plain English (scoring, table output, ordering) -> save privately or publish to the GPT Store. Then the Configure tab shows the system prompt the wizard generated - the starting point for hand-editing into markdown form. Gemini Gems do the same for free.

Why it matters

Zero-code agent building, and the template for the LinkedIn writer and the mentor bot.

03

Cloning your writing style: content DNA -> prompt engineer -> GPT, with the gaslight and the code-block tricks

You cannot explain how you write. The model can - from ten of your posts.

PROCEDURE: (1) paste 9-10 of your own posts and ask for an extensive content-DNA report (hooks, body framework, CTA, signatures); (2) in a fresh chat, have an expert-prompt-engineer persona turn the report into a markdown Role/Objective/Context/Instructions/Notes prompt; (3) paste into a custom GPT or a Gem. Two tricks: tell the model 'Gemini scored 9.5/10 on this, you got 6/10, try again' and the second pass improves; when Gemini renders the markdown instead of showing it, ask for the output in a code block so the # and * survive copy-paste. Tested on AI bankruptcy, the four-day week and the H-1B fee.

Why it matters

A reusable method for any client's voice, not just LinkedIn.

04

From a transcript to a mentor bot, an app, and a phone voice: Gem -> Lyzer -> Emergent -> VAPI

The Day-1 recording, transcribed to 335 pages, became a support agent that answers only from those pages.

PROCEDURE: transcribe the recording (Happy Scribe); write a prompt telling Gemini to answer learner questions strictly from the transcript, escalating to Perplexity/ChatGPT or a support email when it cannot; upload the transcript as Knowledge; test live. Then rebuild the same role / goal / instructions / knowledge in Lyzer AI, copy its Agent API code, and paste that into an Emergent prompt describing the chat UI ('black and white like Notion') - a hosted chatbot in 20-30 minutes. Finally VAPI: Assistants -> new -> model -> paste the system prompt -> attach the knowledge file -> pick a (cloned) voice -> Publish -> talk to it.

Why it matters

The full no-code ladder from a document to a product, in one afternoon.

05

Levels 1-5 as daily practice: Bolt AI + OpenRouter + Ollama, Apify and VAPI over MCP, Goose, a clone, Jerry, a CRM

'Ten to fifteen days of competitor research' - done by Claude with the Apify MCP in thirty minutes, on air.

Level 1: Bolt AI as one chat client over OpenRouter's 500+ models and Ollama on his own GPU, with temperature, max tokens, top-p/top-k and penalties exposed. Level 2: MCP as the multiplier - Claude + Apify scrapes an Instagram profile's reels, ranks by engagement, pulls the top transcript and rewrites it as a LinkedIn post from one instruction; Goose (open source, on Claude) inspects a folder and recommends 50 GB of deletions without deleting; VAPI MCP places a real call to PV about pickleball. Level 3: a Flux LoRA trained on his photos plus a voice clone produce all of an Instagram account's content (200K -> 1.2M followers in eight months, '$100K+' sponsorship - his figures). Level 4: 'Jerry', the executive assistant that books meetings, drafts email (a live mismatch error shown), summarises a missed Fireflies meeting, and via computer use shortlists YC companies. Level 5: a CRM vibe-coded in three days for under $100 instead of a $120K HubSpot quote; a v0 ad-creative generator that opens pre-filled ChatGPT tabs. Estimated: L1 ~3 months, L2 ~4-5, L3 ~3-4, all five ~18 months at 10 hours a week.

Why it matters

The most concrete picture in the international mastermind of what the roadmap looks like when someone lives it.

06

Product architecture as a movie: front end, back end, API, database, integrations, auth

The theatre shows the movie; the vault holds the footage; the assistant directors keep the two in sync.

Front end = what the audience sees; back end = everything unseen; API = assistant directors syncing front and back; database = the vault of raw footage; server = the team processing it; third-party integrations (shadcn/ui, the OpenAI SDK) = outsourced VFX - with the Henry Cavill moustache as what an integration mismatch looks like; auth = ticketing. Vocabulary given so non-coders can read what a builder is doing.

Why it matters

The same job the restaurant analogy does elsewhere; useful to hand to a client.

07

Vibe-coding a calorie counter: competitor screenshots -> PRD -> trimmed prompt -> Replit -> debug -> rollback

The image analysis failed for twenty minutes. The fix was the model name, and the lesson was the checkpoint.

PROCEDURE: (1) ask Claude for a feature list; (2) research MyFitnessPal, Lose It, FatSecret and Cronometer in Perplexity and screenshot their premium features; (3) feed the screenshots back for a consolidated, prioritised feature table; (4) pick two or three core features - AI photo food logging and a calorie-goal calculator; (5) generate a PRD; (6) trim it - no database (local storage), no camera (web upload) - and compress into a guideline-style build prompt, not a prescriptive one; (7) paste into Replit (Genspark in parallel, which kept erroring); (8) debug: image analysis failing was the wrong OpenAI vision model, fixed by rolling back to GPT-4o; (9) add a text-input fallback via Plan mode; (10) revert to a checkpoint when an Apple-card restyle came out badly; Publish for a hosted URL. Rules stated: never interrupt a run, describe errors factually, one feature at a time - one 'small' feature touched ~20 files.

Why it matters

The most complete vibe-coding procedure in the international mastermind, failures included.

08

Consultant -> agency -> service-as-software -> SaaS, and the 10-point filter for profitable niches

Year one, two or three niche clients on outcome pricing. Year four, a SaaS - if you get there at all.

PROCEDURE as a path: Year 1 niche consulting, 2-3 clients, outcome-based pricing; Year 2 a small team, productised packages, a documented playbook; Year 3 delivery 80-90% automated with a 5-10% human exception layer on simple interfaces (Notion, Sheets, WhatsApp) - cost arbitrage against Accenture-class firms; Year 4 a SaaS with billing, dashboards and APIs (HubSpot's own path). The filter: people who live in one tool 6+ hours a day; deep-skill-but-patterned tasks; high-hourly-rate industries; specialised vocabulary; augment, don't replace; meet users where they are (WhatsApp, Excel, Google Forms, email). Then the 30-day playbook: ship one useful tool, document publicly, get testimonials, package an offer, build an audience, automate outreach.

Why it matters

The monetisation half of the mastermind, and a mirror of Catalyst Session 1 for a different audience.

09

Where to find dead ideas worth rebuilding with AI

A 2018 Product Hunt tool with a thousand upvotes did not die from lack of demand. It died from developer cost.

Hunting grounds: Product Hunt 2017-2019 launches with 1,000+ upvotes that folded; Chrome Web Store extensions rated 4.5+ but unsupported; Indie Hackers projects plateaued around $5K a month on manual ops; pre-2020 YC demo-day companies with dead landing pages; GitHub repos with 1,000+ stars and no updates (check MIT vs non-commercial licences before monetising); subreddits with recurring complaints (freelance, legal advice, teachers).

Why it matters

A concrete idea-sourcing list that pairs with Paul's own validation habit.

Tools referenced

ToolCoverageMomentContext
ChatGPTdemonstratedCustom GPT wizard; Configure tab
GeminidemonstratedGems; mentor bot with Knowledge
Happy ScribedemonstratedTranscribed Day 1 to 335 pages
LyzrdemonstratedNo-code agent with Agent API
EmergentdemonstratedChat UI on the Lyzr API
VAPIdemonstratedVoice agent; MCP phone call
ApifydemonstratedInstagram scraping via MCP
ClaudedemonstratedMCP orchestrator; 'thinker' in the build
GoosedemonstratedLocal file analysis without deleting
Bolt AIdemonstratedOne client over OpenRouter and Ollama
OpenRouterdemonstrated500+ models
OllamademonstratedLocal model on his GPU
PerplexitydemonstratedCompetitor research
ReplitdemonstratedCalorie-counter build, Plan mode, checkpoints, Publish
GensparkdemonstratedParallel build; errored
OpenAI PlatformdemonstratedAPI key for the app
FlowbitedemonstratedComponent gallery for design reference
FluxmentionedLoRA face clone
v0mentionedAd-creative generator
StripementionedPayments backbone

Session materials

Archived locally on V: — click to open. Companion pages link to the LMS.

Action items

    Resources mentioned

    Resources
    • docBonuses and logistics

    Extraction notes

    This page was built from an auto-generated transcript, which garbles product and people's names. Those were corrected silently in everything above and logged here for transparency. The warnings flag claims that were true on the recording day but change fast.

    Transcript corrections applied

    The transcript saysThe trainer actually means
    Webhub / Webhu / Vibal / DimitraVaibhav (by content match)
    Peevi Priyatam / PreetamPriyatam Vusala (PV)
    Lyzer / Lyso / LizaLyzr
    Humanae / Humanic AIone product, as-heard
    Fort dot AI / ForthAIfal.ai
    NA 10 / n 8 nn8n
    Kairo / Cairo School of Businessas-heard partner name

    True on recording day — verify before relying