Founder Signal

Inspiration

Building a startup today isn't limited by access to information; it's limited by the ability to connect the right information at the right time. Founders need to constantly monitor competitors, funding rounds, acquisitions, new AI model releases, regulations, and infrastructure changes, yet this information is scattered across countless websites and news articles. Existing AI tools summarize information, but they rarely explain how ecosystem changes actually affect a specific startup.

We wanted to build an AI system that thinks more like a strategist than a search engine. Instead of retrieving documents and generating answers, it constructs a knowledge graph of the startup ecosystem and reasons over relationships between companies, technologies, investors, markets, products, and events. Since this was a Jac hackathon, it was also the perfect opportunity to explore Jac's graph-native programming model, walkers, and byLLM capabilities to build a true multi-agent reasoning system.

What it does

Founder Signal is an AI-powered strategic copilot for seed-stage startup founders.

A founder enters their company name, website, and a short description. The system first determines whether the company already exists in its curated ecosystem graph or whether it should create a dynamic session node for a previously unseen startup. It then discovers relevant ecosystem information from the web, expands the graph with newly discovered companies, technologies, products, investors, markets, and events, and maps the startup into its surrounding ecosystem.

Once the ecosystem has been constructed, founders can ask strategic questions such as:

Which competitors should I monitor?
How should I differentiate my product?
Which technologies are most commonly adopted in my space?
Which investors are aligned with companies like mine?
How do recent acquisitions or product launches affect my strategy?

Instead of producing opaque LLM responses, every recommendation is generated through graph traversal and includes transparent reasoning paths, supporting evidence, confidence scores, and prioritized actions.

How we built it

The project is built entirely around Jac, where graphs are both the data model and the execution model.

The ecosystem is represented as a graph containing companies, products, technologies, markets, customer segments, investors, infrastructure providers, regulations, and events connected through semantic relationships such as operates_in, competes_with, depends_on, funded_by, serves, and affects.

The system is implemented as a pipeline of specialized agents:

Startup Understanding Agent resolves or creates the startup node and enriches missing information.
Graph Expansion Agent plans targeted web searches and uses Firecrawl to discover relevant ecosystem information.
Entity Extraction Agent converts scraped web pages into structured graph nodes and relationships using Jac's byLLM.
Event Discovery & Event Extraction Agents identify acquisitions, funding rounds, partnerships, and product launches from trusted news sources.
Ecosystem Mapping Agent constructs the startup's surrounding neighborhood by traversing graph relationships.
Graph Reasoning Agent uses Jac walkers to perform multi-hop graph traversal and identify evidence-backed reasoning paths.
Strategy Synthesis Agent transforms graph evidence into actionable founder recommendations using byLLM.

The frontend is a lightweight HTML/CSS/JavaScript dashboard with an interactive D3 force-directed graph, allowing founders to explore relationships visually while interacting with the AI copilot.

Challenges we ran into

One of the biggest challenges was deciding how much of the ecosystem should exist before analysis begins.

Our original idea was to create a massive static graph containing thousands of AI companies and relationships. We quickly realized that maintaining such a graph would be impractical during a hackathon and would become outdated quickly.

We instead designed a hybrid architecture with:

a curated base graph for high-quality ecosystem knowledge,
dynamic session graphs that expand using Firecrawl during analysis,
graph-first reasoning that operates on structured relationships instead of raw web documents.

Another challenge was reducing hallucinations. We intentionally separated extraction from reasoning: LLMs only extract structured entities from source material or synthesize recommendations from graph evidence, while the graph itself remains the source of truth.

Finally, designing graph traversal required careful path scoring so that generic hub nodes (for example, "LLMs") would not dominate reasoning paths over more meaningful competitor or market relationships.

Accomplishments that we're proud of

Built a complete multi-agent AI pipeline in Jac.
Combined curated knowledge graphs with dynamically discovered ecosystem information.
Implemented explainable graph reasoning using Jac walkers instead of document-based retrieval.
Integrated Firecrawl for automated ecosystem discovery.
Used Jac's byLLM to produce structured graph updates rather than free-form text.
Built an interactive graph visualization that allows founders to inspect the reasoning behind recommendations.
Created a flexible architecture that supports both curated companies and entirely new startups discovered from the web.

What we learned

This project fundamentally changed how we think about combining LLMs and knowledge graphs.

We learned that graphs are most valuable as a reasoning layer, not simply a storage format. Separating extraction, graph construction, traversal, and synthesis makes the entire system more transparent, modular, and trustworthy.

Working with Jac also demonstrated how naturally graph-native programming maps to multi-agent systems. Walkers provide an intuitive abstraction for autonomous agents that traverse and manipulate graph structures, while byLLM makes it straightforward to integrate structured LLM outputs directly into graph workflows.

Most importantly, we learned that explainability becomes significantly more meaningful when every recommendation can be traced through explicit graph relationships and supporting evidence instead of hidden model reasoning.

What's next for Founder Intelligence Graph

We plan to continue expanding the platform beyond the hackathon.

Future improvements include:

Continuous ecosystem monitoring with scheduled graph updates.
Persistent graph growth across analyses.
Pitch deck ingestion for startup-specific strategy.
Richer investor and accelerator recommendation workflows.
Better graph ranking, confidence estimation, and visualization.
Support for additional startup ecosystems beyond AI.

Our long-term vision is to build an AI strategic copilot that continuously understands the startup ecosystem and helps founders make faster, better-informed decisions through transparent, graph-backed reasoning.