browse pilot

Startup research is fragmented. Analysts jump between spreadsheets, product pages, funding updates, and notes before they can form a clear opinion. We wanted to turn that scattered process into an AI-assisted, transparent workflow where every recommendation can be inspected instead of treated like a black box.

What it does

BrowsePilot is an autonomous startup-scout workspace that turns product data into investment-ready decisions.

It allows a user or AI agent to:

Search and filter startup opportunities by category, price/funding, status, tags, and keywords
Explain why each company matched a scout mission
Score opportunities with a visible Scout Fit score
Compare up to three companies side by side
Visualize market, execution, competition, and regulation risk
Track a watchlist and simulate new company signals
Generate executive market briefs and downloadable investment memos
Run a one-click Judge Mode demo from discovery to decision

How we built it

BrowsePilot is a single-file web application built with vanilla HTML, CSS, and JavaScript. The interface uses Tailwind CSS via CDN.

We implemented browser-native WebMCP tools using the imperative registration pattern:

filter_and_search_data
export_dataset_summary
generate_market_brief

Each tool validates its input, updates the live UI, records activity in an agent observability log, and returns structured output for AI-agent consumption. The app also includes a browser-only fallback so its interactive features still work when WebMCP is unavailable.

Challenges we ran into

WebMCP support depends on the browser environment and secure-context requirements. Loading a local file with file:/// does not expose the WebMCP API, so we designed a graceful fallback mode and tested the project from a local server.

Another challenge was making AI-driven scoring trustworthy. Instead of hiding the reasoning, BrowsePilot shows evidence tags, status signals, fit scores, and risk factors directly in the interface.

Accomplishments that we're proud of

We created more than a data dashboard. BrowsePilot delivers an end-to-end decision workflow:

Discover opportunities
Filter and explain evidence
Compare conviction and risk
Generate an investment-ready memo

We are especially proud of Judge Mode, which demonstrates the complete workflow in under a minute, and the observability panel, which makes agent actions visible and understandable.

What we learned

We learned that an effective AI-agent interface needs more than tool calling. It needs clear state, recoverable errors, visible reasoning, and outputs people can act on.

We also learned how WebMCP can make websites more useful to AI agents by exposing safe, structured browser tools instead of relying only on text or screenshots.

What's next for BrowsePilot

Next, we plan to add live data sources, source provenance and freshness signals, persistent watchlists, richer evidence drawers, and multi-agent diligence handoffs.

Our long-term goal is to make BrowsePilot a trusted research copilot for investors, accelerators, and product strategy teams.