The situation
Analysts at smaller institutions covering US-listed companies need the depth of a large research desk on a fraction of the budget. The raw material is public: 10-K and 10-Q filings, proxy statements, earnings call transcripts and market data. The work is reading all of it, consistently, across several years of history, on every name in the universe, every quarter.
What we built
- Start from a ticker. An analyst enters the ticker of a US-listed company. The platform pulls its 10-K and 10-Q filings from SEC EDGAR, proxy statements and earnings call transcripts, and stores them behind an abstraction that swaps local storage for cloud object storage without touching calling code.
- A financial layer. Five years of prices, statements and ratios from Yahoo Finance, normalized into Postgres, so the research agent reasons over stored numbers rather than re-fetching under load.
- Synthesis across sources. The agent reads the filings, the earnings call transcripts and the market data together, so a memo weighs what management said alongside what the reported numbers show.
- Insights and growth signals. Five-year trends in revenue, margins, cash flow and balance sheet strength, turned into forward-looking signals on growth, risk and capital allocation, each tied back to the source that supports it.
- A memo engine. Investment memos generated against one fixed analytical framework, so two names written up a month apart are comparable.
- A research dashboard. Watchlists, a top-100 universe with benchmark, market overview, a filings browser and rendered memos in one interface.
- Session security. Signed tokens in HTTP-only cookies; every dashboard route is refused without a valid session.
How it is engineered
A single application with one API process, one Celery background worker, one Postgres database and one Redis cache. Long-running ingestion and generation run as queued tasks rather than inside a request, which is what lets a heavy research job survive a page refresh.
Why it matters
Auditable output is the difference between a research tool and a plausible paragraph. Because every claim traces back to a filing, a transcript line or a stored financial, an analyst can check a memo in minutes instead of redoing it.