Inspiration

Before building, we talked to investors about where sourcing breaks down. The common thread was coordination: founder conversations are spread across email, LinkedIn, messaging apps and events, and nobody is sure who on the team already talked to whom or who should follow up.

We interviewed 6 people, including a VC associate at a consumer and CPG fund. She told us her LinkedIn inbox "is just too much," described her exact weekly brief, said she would trust an AI filter, and gave us a real "park until we raise" example. Her advice was to "combine tracking with filtering." That became Last Touch.

What it does

Last Touch gives each person at a VC firm their own agent. It sits next to the tools a fund already uses and does not replace the CRM.

  1. Onboarding interview: your agent asks about your role, thesis, the signal you trust, where you talk to founders, and your hard no's. It builds your scoring criteria from the answers and joins the firm's agent network.
  2. Sourcing: it scans your connected channels, drops spam, vendor pitches and copy-paste messages with a reason, and ranks the remaining founders by your own criteria. Every card shows why.
  3. Agent-to-agent checks: before anyone reaches out, your agent asks your teammates' agents whether they already know the founder. In the demo, Ben's agent replies that Ben met Maya 46 days ago and never followed up, so Ben sends the re-intro and Cara, whose thesis fits better, takes the first call.
  4. Firm view: one screen shows who is talking to which founder and what happens next.
  5. Parked founders: founders parked for timing, for example while the fund is still raising, stay on the list with the reason and the trigger to re-engage.

Follow-ups are drafts. Nothing is sent without the investor's approval.

How we built it

  • Nicrron API for model calls: reading inbound messages, classifying sector and stage, and drafting follow-ups.
  • Roundtable as the agents' shared memory, so each agent can see who on the team already knows a founder.
  • Akta API for company and industry data. Where a live key wasn't connected, the demo shows labeled fixtures.
  • PRISM (prismtrace-sdk) to trace agent runs step by step.
  • Plain code, not AI, for the decisions that matter: thesis fit, warm intro, founder fit and traction, weighted from your onboarding answers, plus a generic-message detector.
  • Stack: Python, Vercel, GitHub.

The model only reads and drafts. It doesn't make the decisions.

Challenges we ran into

  • Privacy: investors talk to founders in the same apps they use with family and friends. We added guardrails so the agent only pulls founder-related conversations, and nothing gets sent without the investor's approval.
  • Changing direction: our first idea didn't survive the investor conversations, so we rebuilt around what we heard.
  • Trust: a filter that silently drops a good founder is worse than no filter. We kept the scoring as fixed rules, showed the breakdown, and logged a reason for every drop.
  • Demo data: we built the demo on a sample fund and inbox, so we could show the full flow without touching anyone's real messages. The comparable companies are real, labeled as fixtures.

Accomplishments that we're proud of

  • A working end-to-end run on 14 inbound messages: 5 dropped with a reason each, the rest ranked by each person's own criteria.
  • Agents resolve who owns each relationship: Ben's agent tells Ava's agent he met Maya 46 days ago with no follow-up, so Ben re-intros and Cara takes the first call.
  • An onboarding interview that turns a few answers into a working agent in under a minute.
  • Parked founders stay visible with the reason and the trigger to re-engage, so nobody gets forgotten.

What we learned

  • The pain we heard was specific and behavioral: weekly briefs, hand-checked senders, founders forgotten after events.
  • Funds do not want another CRM. The product has to be a layer.
  • Trust in an AI filter comes from visible rules and reasons, not from a better model.

What's next for Last Touch

  • Connect to multiple communication channels.
  • Pilot with the funds we interviewed, then test per-seat pricing against what funds pay for CRM seats.
  • Search across the whole team's network ("who do we know working on X?"). One investor told us he does this by hand every time.
  • A small MCP server, so a partner can ask their own AI assistant "which parked founders just raised?"
  • Sync to the CRM a fund already uses, such as Folk or Attio.

Built With

Share this project:

Updates

Submission history