An India-first AI research engine for NSE/BSE equities — index-like returns at roughly half the drawdown.
Maven is an AI market-intelligence workspace for Indian equities. It explains the market the way a good analyst would — mechanisms, not tips. Ask it why banks led this week, what softer crude does to Indian sectors, or how FII flows move the rupee, and it answers with structured reasoning, real data points, and sources.
Behind the chat sit two live, paper-traded strategy engines and a public, auditable track of every trade and price path. The whole thing is built on one core idea, validated across two market eras including the COVID crash:
Index-like returns at roughly half the drawdown. Risk management is the edge.
Maven is a research tool. It is educational, paper-traded, and explicitly not investment advice — it refuses buy/sell/hold calls by design.
- 🧠 Ask Maven — a DeepSeek-powered research pipeline (intent → research → synthesis → validate → score) that grounds every claim in data and refuses advice.
- 📈 Two live engines — Enhanced F+ (vol-adjusted momentum) and Defensive (low-volatility), both paper-traded from ₹10L, side by side.
- 🧾 Every trade, shown — each position is a mechanical decision from real end-of-day NSE/BSE prices, with its price path and the reason it was taken.
- 🏆 Models, ranked — backtested strategies with honest numbers; everything else stays in validation until it earns its place.
- 🎛️ Risk-first by design — drawdown is the headline metric, not return. Graded cash exposure de-risks when the regime turns.
- 🎨 Quiet-luxury UI — near-black canvas, a single emerald accent, editorial serif headlines, fluid Framer Motion.
Note
All performance figures below are backtested (2021–2026), not a live track record, and are shown for education only. See the disclaimer.
"Educational market context for NSE/BSE — what moved and why it matters. Not investment advice."
The chat is a multi-stage research pipeline, not a thin prompt. A request is classified by
intent, a research plan gathers live market data and sources, an answer is synthesised, then
validated against the gathered facts and scored — a weak answer triggers a stricter second pass.
Answers come back structured: a headline, a plain-English summary, DATA / MECHANISM / MACRO /
RISK / TAKEAWAY blocks, citations with timestamps, and follow-up questions. It is powered by
DeepSeek (OpenAI-compatible) and falls back to an offline synthesiser when no key is set, so the
UI never breaks. A guard layer refuses buy/sell/hold requests and explains what Maven can do instead.
The explainer walks through Maven's four layers — Market Mode (what is the Indian market saying?) → Portfolio Mode (what should my portfolio do about it?) → Watchlist (track what matters, personally) → Broker (where insight becomes action) — and the three principles that drive the edge:
- Reads the regime — knows the difference between a healthy market and a falling one.
- Holds quality, diversified — durable businesses, spread across sectors.
- Moves to cash when it matters — the de-risking discipline that halved the drawdown.
Headline: +129.97% (Enhanced F+) vs +82.17% (Nifty 500), 2021–26 — at a lower drawdown (14.05% vs 18.59%). During COVID, F+ fell −27% while the market fell −38%. Backtested, not a live track record.
"Two live books, two engines."
Two strategies run as ₹10L paper books, side by side: Enhanced F+ trades vol-adjusted momentum; Defensive trades low-volatility with sooner, harder de-risking. Each book shows its headline value and alpha vs the Nifty 500 TRI, an equity curve, a graded cash-exposure gauge (100 / 50 / 25% set by the market regime), risk-first stats (max drawdown, Sharpe), and full holdings.
Every position each engine has ever taken — with its price path and the why. The data banner is
live: real prices_eod_adjusted rows from Postgres across the NSE/BSE universe.
"Every trade is a mechanical decision from real end-of-day prices (no discretion) — Enhanced F+ scores on vol-adjusted momentum, Defensive on low volatility."
"Models, ranked. Two strategies are live and fully backtested. The rest are in validation — no numbers until they've earned them."
| Strategy | Total return | Max drawdown | COVID drawdown | For |
|---|---|---|---|---|
| 👑 Quant (Enhanced F+) | +129.97% | 14.05% | ~14% (vs market ~38%) | Growth with crash protection |
| Defensive | +84.61% | 17.63% | ~9% (vs market ~38%) | Capital protection, smoother ride |
| Concentrated | +152.66% | 12.63% | ~21% (vs market ~38%) | Maximum conviction, accepts concentration risk |
Seven more models — Growth, Core, Value, Income, Momentum, Quality, Constrained — are shown with no figures until they pass validation. All numbers backtested 2021–26, not a live track record.
| Layer | Choice |
|---|---|
| Framework | Next.js 14 (App Router) · React 18 · TypeScript |
| Styling | Tailwind CSS · Framer Motion (animation) |
| Charts | Recharts (equity curves, exposure gauges, sparklines) |
| Data | PostgreSQL (Neon-hosted) — real end-of-day NSE/BSE prices |
| AI | DeepSeek (OpenAI-compatible) research pipeline, with offline fallback |
| Type system | Fraunces (display serif) · Hanken Grotesk (UI) · JetBrains Mono (data) |
git clone https://github.com/rxhils/trymaven.git
cd trymaven
cp .env.example .env.local # optional — see below
npm install
npm run dev # http://localhost:3000It runs with zero config. With no database and no API key, Maven serves mock data and the chat
uses its offline synthesiser. To wire the real thing, fill .env.local:
| Variable | What it unlocks |
|---|---|
DATABASE_URL |
Live portfolio / trades data from Postgres (Neon) |
DEEPSEEK_API_KEY |
The DeepSeek-powered chat research pipeline |
TAVILY_/SERPER_/EXA_/BRAVE_API_KEY |
Optional web-search sources for the chat (any one) |
.env.local is gitignored — never commit real keys.
- Palette — canvas
#0a0b0d, panels#111316, ink#e9ebed; one accent, emerald#34d399; rose#fb7185for losses only; muted gold#c9a961for editorial touches. - Type — Fraunces for editorial headlines (used sparingly), Hanken Grotesk for the interface, JetBrains Mono for tabular numbers.
- Motion — scroll-triggered reveals and count-ups via Framer Motion, all respecting
prefers-reduced-motion. - Voice — honesty over hype: mechanism before recommendation, drawdown before return, ₹ lakh/crore and NSE/BSE conventions throughout.
Research tool. Not investment advice. Paper-traded results, not real money. Performance figures are backtested, not a live track record. Built for personal research; not registered with SEBI. Maven does not give buy / sell / hold advice or price targets — use it to understand market dynamics, not as a trading signal.
MIT © Maven · trymaven.in