Yash
    03 / 05·Live·All case studies

    StartupDeckAI

    AI startup analysis and pitch decks, built full stack.

    Live siteDocs
    StartupDeckAI — project preview

    Overview

    StartupDeckAI is an AI platform that turns a startup idea into market analysis, competitor intelligence, a scored GO / PIVOT / KILL verdict, and an investor-ready pitch deck. It is live with 214+ users.

    I built it end to end: the React front end, the API and data model, the analysis engine, the background workers, payments, and the CI/CD pipeline.

    Problem

    Founders juggle spreadsheets, research tabs, and slide templates, and the pitch ends up disconnected from the analysis.

    Generic chatbot output is hard to trust and harder to turn into something investors will read.

    Architecture

    The front end is React and TypeScript on Vite, with TanStack Query for server state, Zustand for client state, Tailwind CSS and Radix UI for the interface, Recharts for charts, and Three.js and Framer Motion for the interactive layer.

    An Express API serves REST and GraphQL on PostgreSQL through Drizzle ORM. Sign-in uses Passport (email and Google OAuth) with sessions in Redis, and payments run on Razorpay.

    The analysis engine runs four LLM stages in parallel and a synthesis stage, then scores eight dimensions with a deterministic formula; every figure carries a source label set in code. Slow generation runs in BullMQ workers on Redis.

    GitHub Actions runs lint, type checks, 3,400+ tests with coverage, an npm audit and Playwright end-to-end tests before every deploy: the front end to Vercel and the API to Render.

    Request path

    1. Web appReact, TanStack Query, Tailwind
    2. APIExpress, REST and GraphQL
    3. PostgreSQLDrizzle ORM

    Background

    1. BullMQ on RedisGeneration jobs
    2. Analysis engineParallel LLM stages, deterministic score
    3. Pitch deckSlides built from the analysis

    Delivery

    1. GitHub ActionsLint, types, 3,400+ tests, Playwright E2E
    2. DeployFront end on Vercel, API on Render
    3. CloudflareCache purged after each release
    Slow AI work runs in workers, and nothing reaches production without passing CI.

    Tradeoffs

    The platform prioritizes a structured idea-to-deck workflow over open-ended conversation.

    Rich 3D visuals cost performance budget, so they are kept progressive and responsive. PDF export is planned rather than shipped in a rough state.

    214+
    Users
    3,400+
    Automated tests
    Idea → Deck
    Core flow

    What I'd do differently

    Additive schema design paid off: pivoting the product from validation to pitch decks did not require a painful migration.

    It also reinforced how SaaS complexity grows once auth, payments, queues, and SEO enter the picture, and why a strict CI gate pays for itself.

    Writing about StartupDeckAI