AI Hedge Fund Analyst Agent - Monitoring and Analysis System Engineer

What I'm building I run a fundamental equity strategy and am building an AI-native hedge fund. The goal, simply put: build an agent that works like a hedge fund analyst — one that monitors everything, knows the investment theses, flags what matters, and does real analytical legwork. The biggest problem in investing is signal-to-noise: enormous amounts of information come out daily about the companies I cover, and the scarce resource is knowing what's worth reading and acting on. I've already built early agent workflows myself (n8n + LLM APIs). I'm now hiring an engineer to build the production version: a system that ingests information continuously across my portfolio and watchlist (20–30 names), filters it against my theses, pushes only what matters, and — for the highest-priority items — does the first round of analytical follow-up itself. This is a scoped contract build, but it must be a real foundation — the first module of a system we keep extending — not a throwaway demo. The build, in two phases Phase 1 — Design + triage pipeline (one contracted scope): Starts with a short design step — architecture, data model, stack recommendation, exact MVP scope — then moves straight into the build: 1. Ingest continuously from defined sources: market/fundamental data APIs (e.g., FMP), SEC EDGAR, news/RSS feeds, transcripts, and my own notes 2. Deduplicate and map new items to companies and themes 3. Score relevance against my thesis cards (core thesis, KPIs, risks, catalysts per name) — cheap-model first pass, stronger model analyzing only what survives 4. Push alerts (email/Slack digest): what happened, why it matters, thesis impact, confidence, source citation 5. Capture my feedback (useful / not useful) and store it so the filter improves 6. Save flagged items and supporting documents in an organized, plain-file structure I can also work with directly in other tools (e.g., Claude) Phase 2 — Analyst layer (on the same foundation): For high-priority alerts, an agentic investigation step: pull related filings and transcripts, check what connected companies (suppliers, customers, competitors) have said, query the data APIs, search prior saved research, and draft a preliminary view with cited evidence — so the output is "here's what happened, here's what I found when I dug in, here's how it bears on the thesis," not just a headline. Every factual claim in any output must link to a source. Non-negotiable. Built to be extended — and operable by me This system is module one of a larger platform (future modules: earnings workflows, catalyst/read across detection, memo drafting, and eventually an orchestrator coordinating them). The Phase 1 data model — companies, thesis cards, documents, alerts, feedback — becomes the shared memory those modules read from. Design accordingly: clean interfaces, modular structure, nothing that boxes us in. I have a CS background and need to be able to operate, tune, and extend the system myself — independent of whether we keep working together (more on that below): • Prompts, thesis cards, source lists, and scoring criteria live in plain editable files — not buried in code • Boring, durable infrastructure: Python, Postgres (or similar), simple scheduled jobs. No exotic frameworks unless clearly justified • A README that takes me from clone to running locally, plus architecture and schema documentation • The bar: I can run it and make a meaningful change (add a source, edit a prompt, adjust scoring) on my own Skills I'm looking for Must have: • Shipped LLM-powered systems in production (not just prototypes or demos) • Strong Python backend skills; comfortable owning a pipeline end to end • RAG/retrieval experience, especially citation grounding and keeping outputs tied to sources • Experience with messy real-world documents (SEC filings, transcripts, news) and data APIs • Agentic workflow experience (tool use, multi-step investigation loops) — via LangGraph, native tool-calling, or well-reasoned custom code • Clean, well-documented work: sane structure, systems that others can build on Strong plus: • Familiarity with equity research workflows (earnings, consensus, catalysts, read-across) • Experience with financial data sources (FMP, EDGAR, transcript providers) Not required: ML research or model-training background. This is applied engineering. How we'd work together • Contract basis to start: Phase 1 (design + build) as one scope with defined milestones and acceptance criteria, then Phase 2 • Work happens in a repo I own from day one; IP assigns to me • I'm deeply involved as the domain expert: thesis cards, source lists, example outputs, fast feedback. You own the engineering • This starts as contract work, but I'm building something much bigger than one module — for the right person, there's a real path to a longer-term role helping build the whole platform How to express interest Send a short note with: 1. 2–3 relevant things you've built, with links where possible — production LLM/RAG/agent work especially 2. High-level thoughts on how you'd build this: rough architecture, stack you'd reach for, what's hard about it, and what you'd build first 3. Your availability, preferred working arrangement, and rough rate