Vigilant Equity Research Agent

Vigilant is an AI-assisted Indian equity research workspace that turns a stock name into a structured, evidence-backed investment view. It is built to answer the question I kept running into while researching public companies: "What would make this stock worth watching, and what would prove me wrong?"

Live app: equity-research-agent.rajatendud.chatgpt.site

What It Does

• Searches Indian listed companies by name or ticker, including aliases such as Asian Paints, Berger Paints, Nerolac, Reliance, TCS, Infosys, and Cupid.
• Builds a buy/watch/avoid-style research view from company fundamentals, price action, valuation signals, risks, and available third-party consensus.
• Separates the output into positives, risks, assumptions, invalidation checks, next checks, valuation read, governance read, financial read, and missing data.
• Uses deterministic fallback logic when AI analysis is unavailable so the app still returns a usable report instead of failing silently.
• Includes a decision journal style interface for scenarios, risk controls, scorecards, performance context, quarterly tracking, and research snapshots.
• Presents the output in a polished liquid-glass UI with responsive layouts for desktop and mobile.

Why I Built This

Most stock research tools either give raw data without a thesis, or a confident recommendation without enough reasoning. Vigilant sits in the middle: it gives a structured view, but keeps the assumptions visible. The goal is not to replace judgment; it is to make judgment harder to fake.

I built it as a full-stack product experiment around AI judgment, financial data workflows, and fast product iteration. It combines my backend/data engineering background with a product surface that is easier for a normal investor to use.

Product Flow

flowchart TD
  A["User enters company"] --> B["Resolve ticker"]
  B --> C["Fetch market and financial signals"]
  C --> D["Build deterministic research base"]
  D --> E["Ask AI for structured narrative"]
  E --> F["Validate JSON schema"]
  F --> G["Render report, risks, scenarios"]

Loading

Tech Stack

Layer	What It Uses
Frontend	Next.js, React, TypeScript, custom CSS
Runtime	Vinext on OpenAI Sites / Cloudflare-style worker runtime
API	Next server routes under app/api
AI	NVIDIA-hosted model integration with strict structured-output validation
Data	Screener company lookup, Yahoo Finance-style consensus/market endpoints, deterministic local analysis
Storage	Drizzle/D1-ready schema and migrations
Testing	Node test runner, build validation, structured-output parser tests

Architecture

The main analysis endpoint is app/api/analyze/route.ts. It performs ticker resolution, market-data collection, deterministic analysis, optional AI enrichment, schema validation, and fallback report generation.

The AI response is intentionally constrained by app/api/analyze/structured-output.ts. This avoids a common failure mode in AI products: the model writes something that looks plausible but breaks the UI or quietly omits important fields.

Key Features

• Structured verdicts: Potentially investable, Wait / watchlist, Avoid, or Insufficient evidence.
• Scenario lens: cautious, base, and upside framing instead of a single overconfident target.
• Risk-first interface: risks, invalidation triggers, and missing data are kept visible.
• Third-party comparison: checks available analyst consensus and shows divergence when usable.
• Fallback analysis: deterministic report generation keeps the product useful during provider/API failures.
• Roadmap-ready product surface: research timeline, snapshots, quarterly tracker, scorecard, and decision-journal concepts are built into the UI direction.

Local Setup

Prerequisites:

• Node.js >=22.13.0
• Linux shell environment with bash, curl, flock, and GNU timeout

Install dependencies:

npm run install:ci

Run locally:

npm run dev

Build and validate:

npm run build

Run tests:

npm test

Check that GitHub/local HEAD matches the deployed live app:

npm run check:live-sync

The live app exposes /api/version, generated at build time. GitHub Actions runs the same sync check hourly and on demand so deployment drift is visible quickly. If it fails, build, save, and deploy the latest tested commit to Sites before calling the release complete.

Environment Variables

The app can run deterministic analysis without AI, but AI narrative generation expects:

NVIDIA_API_KEY=your_key_here

Do not commit real API keys. The repository only references environment variables.

Important Files

Path	Purpose
app/page.tsx	Main product UI
app/api/analyze/route.ts	Core research API and analysis orchestration
app/api/analyze/structured-output.ts	AI JSON schema and parser
app/api/quote/route.ts	Quote/data route
app/api/research/route.ts	Saved research route
app/api/snapshots/route.ts	Snapshot route
db/schema.ts	Drizzle/D1 schema
tests/nvidia-structured-output.test.ts	AI schema/parser tests
tests/rendered-html.test.mjs	Rendered metadata/build sanity test

Validation Status

Current validation:

npm test

Result:

• Production build completed successfully.
• Sites artifact validation passed.
• 5 tests passed.
• 0 tests failed.
• Live sync check available via npm run check:live-sync.

Latest Product Iteration

• Stabilized Watchtower evidence review by replacing DOM-level click handlers with React state.
• Fresh evidence refreshes now save dated snapshots, making repeated scans comparable over time.
• Added a clear Evidence diff action beside Review so thesis review and source-change review are separate user flows.
• Preserved fresh-clone build reliability by ensuring project shell scripts are executable.
• Added the first Portfolio command centre pass inside Watchtower with owned-idea return read, review burden, and concentration prompts.

Roadmap

• Add richer primary-source evidence from annual reports, exchange filings, and earnings calls.
• Add a research timeline with dated thesis changes.
• Add user-owned watchlists and decision journals.
• Improve peer comparison and sector-level views.
• Add confidence-stability tracking across repeated runs.
• Expand coverage for small-cap and less frequently covered Indian stocks.

Disclaimer

Vigilant is a research and decision-support tool, not financial advice. It can help structure a thesis, surface risks, and organize evidence, but investment decisions still require independent verification.