KS
← Back to projects
CompletedFeb 28, 2026

QUANTBRIEF: AI-POWERED REAL-TIME MARKET INTELLIGENCE AGENT

Multi-agent financial intelligence platform built at Mistral AI Hackathon — 5-agent pipeline screening 30+ articles in <50ms each, analyzing full SEC filings in 256K context, and delivering multilingual audio briefings.

lines 13.5K+
pythonmistral-aifastapireactmulti-agentquantitative-financenlpelevenlabs
View on GitHub →

Overview

QuantBrief is a multi-agent market intelligence platform that acts as a personal Chief Investment Officer for retail investors. It monitors markets, SEC filings, and news sources automatically — then delivers synthesized intelligence briefs on a user-defined schedule.

Built in 48 hours at the Mistral AI Worldwide Hackathon 2026 (Anything Goes track). The system leverages 4 specialized Mistral models, each chosen for a specific strength in the analysis pipeline.

The core problem: 150+ million retail investors face 300+ financial articles per hour and complex 80–120 page SEC filings, yet Bloomberg Terminal costs $25,200/year. QuantBrief democratizes this intelligence.

Architecture

┌─────────────────────────────────────────────────────────┐
│               DATA INGESTION (Free APIs)                │
│   SEC EDGAR · yfinance · RSS feeds · FRED macro data    │
└────────────────────────┬────────────────────────────────┘
                         ▼
┌─────────────────────────────────────────────────────────┐
│          5-AGENT ORCHESTRATED PIPELINE                   │
│                                                         │
│  ① News Screener     (Ministral 3B)     ~50ms/article  │
│  ② Filing Analyst    (Mistral Large 3)  256K context    │
│  ③ Reasoning Engine  (Magistral Medium) CoT analysis    │
│  ④ Synthesizer       (Mistral Large 3)  Cross-source    │
│  ⑤ Voice Agent       (ElevenLabs)       5 languages     │
└────────────────────────┬────────────────────────────────┘
                         ▼
┌──────────────────────┬──────────────────────────────────┐
│   REACT DASHBOARD    │         REST API (25+)           │
│   Real-time WS       │   JWT auth · Rate limiting       │
│   Charts · Audio     │   PostgreSQL · Redis cache        │
└──────────────────────┴──────────────────────────────────┘

Multi-Agent Pipeline

Each agent uses a different Mistral model optimized for its role:

  • News Screener (Ministral 3B) — High-throughput batch screening of 30+ articles, JSON-structured relevance and sentiment scoring at ~50ms per article
  • Filing Analyst (Mistral Large 3, 256K) — Ingests entire 80–120 page SEC filings (10-K, 10-Q, 8-K) in a single context window — no RAG, no chunking, full document understanding
  • Reasoning Engine (Magistral Medium) — Chain-of-thought portfolio impact assessment with confidence scores
  • Synthesizer (Mistral Large 3) — Cross-source correlation into executive briefs with actionable items
  • Voice Agent (ElevenLabs) — Professional multilingual audio briefings in 5 languages (EN/FR/DE/PL/ES)

Key Features

  • Scheduled brief generation — every 4h, daily, or weekly on user-defined UTC time
  • Real-time market data with technical analysis (RSI, MACD, Bollinger Bands)
  • Interactive candlestick charts (Recharts + Lightweight Charts)
  • Earnings call transcription and analysis via Mistral multimodal
  • Portfolio and watchlist management with per-ticker news sentiment
  • Full JWT authentication with access/refresh token flow
  • WebSocket pipeline progress (real-time stage + percentage)
  • 33 automated tests with CI/CD via GitHub Actions

Tech Stack

Python, FastAPI, SQLAlchemy, PostgreSQL, Redis, Mistral SDK, ElevenLabs, APScheduler, React 19, TypeScript, Vite, Zustand, Recharts, Tailwind CSS, Docker, Vercel, Railway