ForgeAlpha - The Autonomous Hedge Fund Analyst

• 

  ForgeAlpha

Inspiration

Retail traders have Robinhood. Hedge funds have Bloomberg terminals, quant pipelines, and analyst desks. We asked: what if one autonomous agent could close that gap — run a real multi-factor pipeline, surface the reasoning transparently, and take action — all before a human finishes their coffee?

What it does

ForgeAlpha ingests 226 stocks across 4 live data sources (Yahoo Finance, X/Twitter, Reddit/ApeWisdom, geopolitical feeds), runs a 7-stage weighted scoring pipeline, and ranks the top 5 BUY recommendations with full rationale. Results surface in a dark Bloomberg-terminal-style Streamlit dashboard where the analyst can tune 7 parameters and watch rankings recalculate live. One click publishes the call to Telegram — a real action on the open web.

How we built it

Stack: Python, Streamlit, Composio, TrueFoundry, Airbyte. We built a worker pipeline (yfinance → X sentiment → Reddit retail → geopolitical → scoring → ETF scanning → recommendations) and wrapped it in a hackathon dashboard. Composio powers the Telegram publish action. TrueFoundry routes AI inference with observability. Airbyte grounds the pipeline in external data via source connectors. All data is real — no mocks. 47K characters of live market data captured from today's run.

Challenges we ran into

Time. 5 hours to integrate 3 sponsor tools, build a Bloomberg-grade dashboard, and make it production-quality. Dashboard polish under time pressure — Streamlit's layout quirks required creative use of columns and custom CSS.

Accomplishments that we're proud of

Real pipeline, real data, real action. Not a demo with mock JSON — 211 stocks scored using actual market data with a verifiable weighted-average formula. The 6-stage stock recommendation funnel (Universe → Market → X → Reddit → Scoring → Top Picks) makes the invisible visible. Seven tunable sliders recalculate live using the real formula, not a toy. The agent published to Telegram from the dashboard — a genuine action on a real platform. And we shipped in 3 hours by reusing mature infrastructure rather than rebuilding.

What we learned

Prebuilt skills help in hermes. We focused the 3 hackathon hours on integration, dashboard, and the agent action layer. Three sponsor tools are more convincing when each does a distinct, visible job — Composio (action), TrueFoundry (AI gateway), Airbyte (data layer) — rather than overlapping.

What's next for ForgeAlpha

Fidelity OAuth integration so the agent can execute trades, not just recommend them. And a natural-language query layer — "show me why COHR dropped from #1 to #4" — powered by the reasoning chain the pipeline already generates.

Built With

• airbyte-cdk
• altair
• composio
• feedparser
• openai
• pandas
• playwright
• protobuf
• pyarrow
• requests
• streamlit
• websockets
• yfinance