A people-expert engine for a deal team.
Atlas maps the human network behind an investment thesis. Give it a thesis in plain language and it returns a verified map of the founders, advisors, bankers, lawyers, and peer-fund GPs who shape that space, along with the companies they connect to. The site currently covers 16 themes.
What it does
Atlas turns a long, fragmented research process into three motions a deal-team member can complete in minutes.
- Prep for a call. Every expert ships with a structured brief: background, what they care about, sensitivities, three specific questions to ask.
- Send outreach. A drafted subject line and email body sit on every expert page. Edit, copy, send.
- Build a shortlist. Pick experts across themes. Atlas generates a one-page memo with everyone's why-they-matter and the companies that link them.
What's in it
Coverage spans 5 archetypes per theme: founder, advisor, banker, lawyer, peer-fund GP. Companies are derived from the relationships experts actually have, ranked by edge depth.
How it works
A thesis is compiled into a research theme with an archetype budget, then a multi-stage agent pipeline scouts candidates, enriches each one against public sources in parallel batches, resolves duplicate identities across themes, and derives companies from the relationship graph the experts actually produce rather than from a second search.
Everything after that is deterministic. The verification layer calls no language model: it resolves each cited URL under rotating browser signatures, parses schema.org records, extracts page titles, matches names and organisations positionally, falls back to dated Internet Archive snapshots when a link has rotted, and grades the result on a five-tier evidence ladder. Results are written to a content-addressed evidence store, so re-auditing the whole corpus is a local operation measured in seconds and the audit can run continuously instead of once.
What ships is the artifact that survives: fully static, with no database, no login, and no model call at request time.
How it's honest
Every expert and every company carries at least one public source URL that resolved and named them at verification time. Records that could not be grounded that way were dropped rather than fabricated, and where a theme has an archetype gap Atlas reports the gap instead of padding it with low-confidence entries.
The rejects are published too. The verification page carries the full audit: the four-way outcome partition for every source link, the evidence grade behind every record, freshness and departure telemetry, and each record that no longer clears the bar, named rather than quietly deleted.