KSVenture Group
Portfolio register
CompletedAs of

QuantBrief

Multi-agent market-briefing system, built at a 48-hour hackathon

QuantBrief is a multi-agent system that screens market news, analyses SEC filings and turns them into scheduled written and audio briefings. Kacper built it in 48 hours at the Mistral AI Worldwide Hackathon 2026 as a full-stack web application with automated tests. It is a finished hackathon build that has not been developed since 1 March 2026.

Key points

  • Built by Kacper Saks in 48 hours at the Mistral AI Worldwide Hackathon 2026, Anything Goes track
  • Five-agent pipeline on Ministral 3B, Mistral Large 3 and Magistral Medium, with ElevenLabs for audio
  • SEC filings go to Mistral Large 3 (256K-token context) in a single call, without chunking or RAG
  • Full stack: FastAPI (25+ endpoints), React 19 dashboard, JWT auth, PostgreSQL, Redis, 33 tests

What it is

QuantBrief is a multi-agent market-briefing system designed for retail investors. It pulls news, SEC filings, market prices and macro data from free data sources (SEC EDGAR, yfinance, RSS feeds and FRED), runs them through a five-stage AI pipeline and produces a written brief with an audio version. Kacper Saks built it over 48 hours, from 28 February to 1 March 2026, at the Mistral AI Worldwide Hackathon 2026 (Anything Goes track), as a full-stack web application with login, a dashboard and scheduled briefs.

Five-agent pipeline

Each stage uses a model matched to its job rather than one model for everything. A small model screens news in batches and returns relevance and sentiment scores as structured JSON; a long-context model analyses filings; a reasoning model assesses step by step how events affect the portfolio and attaches confidence scores; Mistral Large 3 then merges all sources into a brief with action items, which ElevenLabs reads out in one of five languages.

  • News screening: Ministral 3B
  • Filing analysis: Mistral Large 3 (256K context)
  • Portfolio-impact reasoning: Magistral Medium
  • Brief synthesis: Mistral Large 3
  • Voice: ElevenLabs, EN/FR/DE/PL/ES

Filings in a single call

The central design choice was to send the text of an SEC filing (10-K, 10-Q or 8-K) to Mistral Large 3 in a single call, relying on its 256K-token context window, instead of splitting the document into chunks and retrieving fragments with RAG. The aim is to keep cross-references between sections intact, for example risk factors that refer to revenue figures and forward guidance. In the hackathon code the extracted filing text is capped at 200,000 characters, so longer filings are truncated rather than split. Uploaded earnings-call recordings are transcribed, and Mistral Large 3 extracts financial highlights, forward guidance, risk factors and Q&A insights from the transcript.

Built in 48 hours

The hackathon build covers both frontend and backend. The backend is a FastAPI service with more than 25 endpoints, JWT authentication with bcrypt password hashing, PostgreSQL, Redis caching, per-endpoint rate limiting and security headers. The React 19 and TypeScript dashboard shows pipeline progress live over WebSocket, interactive price charts and technical indicators (RSI, MACD, Bollinger Bands), and its interface is available in five languages. Briefs can be scheduled every 4 hours, daily or weekly, with schedules stored in PostgreSQL. The backend has 33 automated tests (pytest), and the repository includes a GitHub Actions CI workflow.

Status

QuantBrief is a completed hackathon project. Its last commit is dated 1 March 2026, and it has not been developed since. The source code is public on GitHub. The demo ran on a temporary URL set up for the hackathon, and its backend no longer responds. For KS Venture Group, the project is a record of what Kacper shipped under a hard 48-hour deadline and of the approach behind it: match each model to its task, and give a long-context model the document in a single call rather than retrieved fragments.

Record

Status
Completed
Stage
Hackathon build, February 2026
Sector
FinTech / AI Agents
As of

Key figures

Build time
48 h
Pipeline agents
5
API endpoints
25+
Automated tests
33

Topics

AI Agents, Mistral AI, FinTech, Hackathon

Further reading

Questions about this project?

Contact the founders