How it works
Exa as your AI research analyst
Exa Vantage is built for the analyst or associate who assembles research. Pick a desk (Investment Bank, Private Equity, or Venture Capital) and the report is tailored to that workflow: IPO comparables, acquisition targets, or a competitive landscape. It compresses the slow first 80% of the work without replacing the analyst's judgment.
Pick a desk, route the request
Choose Investment Bank, Private Equity, or Venture Capital, then type a company or an industry. Each desk tailors the whole report.
Gemini 3.1 Flash Lite classifies the input as company vs industry and resolves a company's domain. The desk sets the lens (IPO comps / buyout targets / landscape).
Discover, then filter for relevance
We find the companies that matter, including ones the big databases have not catalogued, and drop the look-alikes that only share a name.
Exa findSimilar (company) plus a semantic search built from the seed's real business; a Gemini relevance gate removes name-collisions, judging by what each company does.
Surface what's emerging
A deep-research pass looks for newer, lower-profile companies the structured databases lag on.
Exa Agent deep research (effort-dialed) finds under-the-radar names; falls back to a freshness-tuned Exa search.
Cluster, profile, and quantify
We group the companies into sub-segments, write a tear sheet for each, and pull the numbers: funding, founded year, stage, headcount, HQ.
A Gemini pass builds the clusters and tear sheets. A per-company Exa search pulls page text, and one batched Gemini pass extracts quant from evidence only (blank when unknown). No invented figures.
Catch recent signals
Each company gets a fresh signal: funding, a launch, a hire.
The same per-company Exa search surfaces the freshest result with content, streamed in with bounded concurrency.
Synthesize into a branded deck
We write the memo and render everything as a page-by-page slide deck in your firm's brand, shareable by link and exportable to PDF.
A Gemini synthesis call writes the closing memo; the deck saves to Firestore for a shareable /r/[id] page and prints one slide per page.
What it runs on
- Exa — neural search, findSimilar, content text, and Exa Agent deep research
- Gemini 3.1 Flash Lite — routing, relevance, clustering, quant, synthesis (seed 7, JSON)
- Next.js on Vercel — App Router, NDJSON streaming route
- Firestore — saved reports for shareable links
Quantitative figures (funding, founded year, headcount, HQ) are estimated from public web sources and left blank when not evident, never invented. Exa Vantage wins on discovery of the unknown, freshness from unstructured sources, and what a company actually does.