BizMatch

• 

  Matching Small Businesses using K2 Think V2

Inspiration

Small businesses are full of underutilized resources, including idle equipment, spare capacity, and niche expertise, while neighboring businesses are actively searching for exactly those things. The problem isn't availability; it's discovery. Traditional networks rely on word-of-mouth or generic directories that match on keywords, not fit. We built BizMatch to change that: an AI-powered layer automates the process of business relationships. Small businesses build each other up by building mutually beneficial partnerships.

What it does

BizMatch connects small businesses that have specific needs with other small businesses willing and able to fulfill them. Describe what your business is looking for (web development, catering, logistics, legal, etc.) or offers, and get matched via the K2 Think V2 LLM to candidate businesses based on capability, compatibility, geographic proximity, and reputation, then returns a fit percentage. See matches plotted geographically with proximity indicators, automating the process of forming business connections for resource movement.

How we built it

Our frontend was built using React 18, Vite, TailwindCSS, and Supabase Auth. As for our Backend API, we used Node.js, Express, TypeScript, and Supabase (PostgreSQL + PostGIS). Python, FastAPI, and K2 Think V2 LLM API powered our AI functionality, and Docker facilitated the containerization of our backend.

Challenges we ran into

A purely geographic rank buried high-reputation remote businesses; a purely reputation-based rank ignored local fit. Thus, we had to mix weightings and factors, which K2 Think V2's complex reasoning enabled us to perform.

Accomplishments that we're proud of

Our design enables users to fluidly manage multiple businesses while coordinating resources provided and needs to engage in beneficial interactions with other businesses.

What we learned

LLMs are excellent at semantic capability matching but need structural scaffolding (strict JSON prompts, defensive parsers, fallback scores) to be reliable in a production pipeline.

What's next for BizMatch

We plan to increase the number of agents in our workflow, automating AI-drafted outreach emails, meeting scheduling, and negotiation assistance, which all require explicit user approval before any action is taken. Also, we aim to capture match acceptance/rejection signals and use them to tune the K2 Think V2 scoring weights over time, moving toward personalized ranking per business.

Built With

• docker
• fastapi
• k2-think-v2
• nextjs
• node.js
• python
• react
• supabase
• tailwind
• typescript
• uvicorn
• vite