TickerStats
AI-powered equity research SaaS that compares tickers, generates pitch decks, runs DCF valuation, and exports to PPTX/PDF/XLSX.
Tech Stack
Tags
Key Outcomes
- Built a 30-ticker relative valuation engine with caching and concurrent data fetch
- Shipped schema-validated LLM deck generation with modular, testable section pipelines
- Implemented tiered billing with Auth0 JWT claims + Stripe and tracked usage quotas in Postgres
- Delivered one-click exports to CSV, XLSX, PDF, and PPTX from server-side generation
Overview
TickerStats is a full-stack SaaS platform for student investment clubs and retail analysts who want a faster way to do equity research without Bloomberg. The product focuses on two things: comparing companies quickly, and turning research into a real deliverable like a pitch deck or exportable model.
What It Does
- Relative valuation and performance comparisons for up to 30 tickers at a time (fundamentals plus return and risk metrics).
- AI pitch deck generator that outputs structured sections like company overview, thesis, risks, SWOT, and valuation.
- Deterministic DCF model with sensitivity analysis to produce a fair value target price.
- Export pipeline for CSV, Excel, PDF, and PowerPoint.
Technical Architecture
The system is split into a React + TypeScript client and a FastAPI backend.
Data is pulled through yfinance and shaped into consistent metric objects for the UI. For large comparisons, requests are parallelized and cached using an in-memory TTL cache to avoid repeated fetches.
The deck generator uses a modular "section spec" design. Each section has its own prompt builder, JSON schema, and post-processor. LLM outputs are schema-validated (Pydantic v2 structured outputs) before they are stored or exported.
Auth and billing are enforced via Auth0 (JWT claims) and Stripe subscription tiers. Usage quotas are tracked per user in PostgreSQL.
The key engineering goal was reliability: schema-validated LLM outputs and deterministic post-processing so generated decks do not break when models respond unpredictably.
Key Challenges
Making LLM outputs safe enough to export
Deck sections are generated as structured JSON and validated end-to-end. If a section fails validation, the system can retry or fall back without corrupting saved work.
Keeping 30-ticker comparisons fast
yfinance calls were parallelized and cached so repeated comparisons do not re-hit the same external data on every request.
Results
TickerStats is designed around a real workflow: research faster, generate a deck, and export the output for meetings and presentations. The platform combines data, modeling, and deliverables in one place instead of spreading the workflow across multiple tools.