MarketMind Hub
A white-label financial intelligence platform for Indian fintech — a RAG-driven market-data engine with three public faces: a marketing site, a Model Context Protocol (MCP) server, and a developer portal with API keys and billing. This page is the overview; each face has its own deep-dive.
Overview
MarketMind Hub is a white-label, AI-driven financial intelligence platform aimed at Indian fintech startups and trading tooling. It exposes real-time market data (market status, trading holidays, T+1 settlement dates, weekly and monthly F&O expiry, lot sizes), ticker intelligence for 2,359 NSE/BSE names, RAG-grounded financial context, and ML price signals — through two public integration channels: a REST API and a Model Context Protocol server. One API key unlocks both.
I build and operate the whole system end to end: the data pipeline, the models, the API backend, the billing, and the three public websites. Each site targets a different user — the marketing site sells the idea, the MCP site converts AI-agent developers, and the developer portal runs the account lifecycle and revenue. This page covers the platform and the strategy; the three links below go deep on each site.
The platform
- Data pipeline: parsers ingest from NSE, BSE, RBI, SEC filings, RSS, and analyst-ratings sources; an event converter normalizes them into financial documents; a RAG uploader ingests 16K+ documents into the vector index (the live site's own figure).
- Retrieval: ONNX-accelerated embeddings plus Redis HNSW vector search power the
rag-querypath, with ticker metadata cached in Cloudflare R2 + KV for the MCP data tools. - Inference: a fine-tuned Mistral-7B for ticker extraction (70K+ training rows) and Chronos-Bolt, a zero-shot time-series model that returns 7-day price forecasts with a median plus p10–p90 confidence band.
- API surface:
api.marketmind-hub.inhosts auth (signup/login/logout + Google OAuth), developer profile, API-key lifecycle, quota tracking, payment init/verify, and the marketplace endpoints — ticker-extract, price-predict, rag-retrieve. - Delivery: REST API and MCP forward, plus a Telegram bot and a Discord bot on the side — with an admin app in Flutter and a separate Render-hosted service for contact and enterprise enquiries.
Subdomain strategy
Four subdomains, each with one job, sharing one brand:
marketmind-hub.in— the front door (GitHub Pages). Explains the product, shows the one-key / two-channel story, funnels visitors toward docs and signup.mcp.marketmind-hub.in— the agent developer's entry point (Cloudflare Worker). The same worker serves both the MCP protocol endpoint and its onboarding site.developers.marketmind-hub.in— the account seat (Vercel). API keys, dashboard, quota, docs, pricing, and Razorpay checkout.api.marketmind-hub.in— the money path. Every one of the three sites routes its authentication and data calls here.
The public sites are deliberately thin static front-ends. Each talks to the same backend, and config-driven linking keeps the whole system consistent — a single config file on the marketing site points every link at the right subdomain.
The three case studies
Each public site gets a full write-up — audience, architecture, systems, and the honest gaps I'd ship differently if I were doing it again.
Main website
The top-of-funnel marketing site on GitHub Pages — light design system, config-driven links, one-key messaging.
Read case studyMCP site
A Cloudflare Worker co-hosting an authless MCP server and the static site that teaches you how to connect.
Read case studyDeveloper portal
The dark console on Vercel — API keys, quotas, API reference, and real Razorpay billing.
Read case studyStatus & roadmap
Live today: three marketplace APIs (ticker-extract, price-predict, rag-retrieve) behind key-based auth with plan-based quotas; an authless 12-tool MCP server; a three-subdomain funnel running across GitHub Pages, Cloudflare, and Vercel; and a working Razorpay checkout with credit-weighted plans.
Publicly marked coming soon: news sentiment scoring, earnings-call intelligence, and a real-time market alerts WebSocket stream. The detail pages below are the full story — including the parts that are still rough.