Inspiration

What it does

How we built it

Challenges we ran into

Accomplishments that we're proud of

Project story: VC Startup Finder

Hackathon track: Track 1 — Sourcing

Team: Collaborative project

Why we built it

A promising company is only the beginning of an investment conversation. Venture capitalists still need to understand how it fits their strategy, which facts support that assessment, what remains unknown, and why it deserves a meeting.

We built VC Startup Finder to connect those questions in one accessible workspace. Investors should be able to describe their investment thesis in everyday language and move through research, evaluation, and collaboration without learning technical tools or coordinating agents themselves.

What the app does

Investors define their thesis through conversation or guided controls, review the interpreted criteria, and submit a proposal for owner approval. Changes require a reason and preserve earlier versions, so the team can understand how its investment strategy evolved.

The live demo searches real companies through akta.pro by name or website. Investors can inspect company identity, website, category, and status, compare available evidence with their thesis, prepare a cited brief, and record a decision. Missing financial or funding information stays unknown. Company websites help distinguish businesses with the same name.

A chat guide walks users through the workflow with suggested actions, simple typed requests, and optional browser speech input. For example, an investor can say “Find Clay,” confirm the search, review the company at clay.com, and ask what to do next. The guide also carries missing-input clarifications into thesis interpretation.

The return workbench exposes assumptions, formulas, ownership changes, dated cash flows, and gross return scenarios. Teams can save versions with their reasoning and share permitted materials through scoped, expiring, revocable access. The responsive PWA supports desktop and mobile use.

How we used Nicrron and the model

Nicrron provides the model route for natural-language thesis interpretation. We selected openai/gpt-6-luna after comparing candidates on a versioned synthetic corpus, retaining failed attempts and recording validation, latency, usage, and estimated cost.

The model proposes an editable thesis; it does not publish the mandate. Structured validation, explicit provider routing, cost limits, and human confirmation constrain its role. General workflow guidance follows application-defined steps, while return calculations use deterministic Python logic. The current evaluation qualifies synthetic preview use; broader investor-rated model validation remains ahead.

How we used Roundtable

Roundtable coordinates scoped evidence and brief agents through versioned files and task handoffs. We verified a live synthetic two-agent workflow, exact source-version references, repeatable replay, and denial of a private-file read by another bot.

Real-company lookups remain within the application unless a pilot firm is explicitly bound for Roundtable processing. This keeps the demonstrated synthetic coordination separate from unapproved transfer of firm or provider data.

How we built it

The application uses Python and FastAPI, Supabase Postgres and Auth, and a JavaScript PWA deployed on Render. Supabase CLI migrations manage the database. Authorization checks enforce firm membership and object-level sharing, while versioned records preserve thesis, scenario, and decision history. Browser checks verify the hosted investment journey and chat guide, including mobile layout and offline clearing.

What comes next

Our next steps are deeper thesis-driven discovery and company enrichment, integration of the teammate's advisory model, and investor review of sourcing quality. Current real-company lookup does not establish fundraising intent or provide complete diligence. Return scenarios are illustrative ordinary-equity calculations; preferences, fees, carry, and net LP reporting require further modeling.

Our goal is to help investors reach a better-supported meeting decision, with the evidence and uncertainty visible throughout.

Live demo: VC Startup Finder

Presentation: Track 1 Sourcing deck

Source code: ithllc/VC_Startup_Finder (private repository)