Liubov
← All work

Relationship intelligence

How close is a venture firm to each person in its network, and how does it know? Built with AlleyCorp for its Deep Tech portfolio.

Partner
AlleyCorp
My part
Database schema, enrichment, MCP server
When
2026, team of 3
Stack
Next.js, TypeScript, Python, MCP, LLMs

The problem

A venture firm’s network is its most valuable asset, and it mostly lives in people’s heads. When a portfolio company needs an intro, the question is always the same: who do we actually know there, and how well?

We worked directly with AlleyCorp’s partners to turn that question into a product.

Warmth you can explain

Every relationship gets a tier: Hot, Warm, Stale or Cold. The tier comes from specific signals, and each one can be shown to the person looking at it:

  • Co-investment history
  • LinkedIn connections
  • Event attendance, pulled from event platforms like Luma and Swoogo
  • Public press
Evidence, not vibesSignals become a tier, but the reasons always travel with it.
SignalsFour kindsCo-investments, connections, events and press
RulesWeight + decayStronger evidence counts more; old evidence fades
Explainable tierHot → ColdEvery rating shows the evidence behind it

Signals fade over time. Meeting someone at an event two years ago counts for less than co-investing with them last quarter. A partner can always see why a relationship is rated the way it is. That was the point.

Where the AI is

CodeSignals and decayCollected and weighted by rules, faded over time.
CodeWarmth tierHot, Warm, Stale or Cold, with the reasons attached.
AIEnrichmentLLMs and web search fill in investor profiles.
AIPlain-language search“Who do we know at climate hardware startups?”
PersonThe introPartners decide who to reach out to.

The rating itself never comes from a model. AI helps fill in the picture and makes it easy to search, but the answer to “how well do we know them” is always traceable.

My part

I owned the database schema, the backend enrichment and the MCP server, which lets AI assistants query the relationship data directly. Access is authenticated only, through deep links, with no anonymous access. A daily digest brings the most useful changes to partners without them having to open anything.

Outcome

DeliveredWorking partner demo

Presented at Demo Day in June 2026 with the database, enrichment and authenticated MCP access in place.

Not claimedNo adoption metric yet

The case study reports what the team built and learned, without presenting a demo as measured business impact.

My biggest lesson: for a tool like this, being able to explain a rating isn’t a nice extra. It’s what makes people willing to trust it.