Guild: The Smart Artisan Network

Inspiration

In Nigeria and across the globe, the informal service sector (plumbers, electricians, carpenters) is massive yet heavily fragmented. It relies almost entirely on word-of-mouth, creating a massive "trust deficit." Clients struggle to find reliable, verified artisans, and skilled pros struggle to find consistent, well-paying work. We realized that by digitizing this ecosystem, we could provide stable income for thousands of workers while giving clients peace of mind.

What it does

Guild is a two-sided, smart marketplace for the informal economy.

• The Smart-Match AI: Instead of clients manually searching, they simply type their problem naturally (e.g., "My ceiling has a wet patch"). Our AI parses the technical requirements, categorizes the urgency, and automatically routes the job to the top verified Pros.
• The Trust Engine: Once a job is agreed upon, Guild uses a secure Escrow Payment System. The client's funds are held securely while the Pro works, and only released once the job is marked complete.
• Instant Handshake: We built a fully real-time negotiation chat room where Clients and Pros can finalize terms instantly.

How we built it

We built the platform using Next.js 15 (App Router) leveraging Server Actions for a lightning-fast frontend, styled with Tailwind CSS.

For our backend, we used Supabase (PostgreSQL) to handle authentication, Row Level Security (RLS), and database management. We specifically utilized Supabase Realtime (WebSockets) to power our instant chat infrastructure.

For the AI matching engine, we used the OpenAI SDK connected to the Featherless.ai serverless inference platform. This allowed us to bypass Vercel's heavy Python hosting limits and use powerful open-source models (like Meta Llama 3.1) entirely server-side via fast API calls. Finally, we integrated Stripe to handle the complex Escrow payment flow.

Challenges we ran into

Building a true two-sided marketplace is incredibly complex. Managing the state of the "Escrow Handshake" (Pending -> Active -> Funded -> Awaiting Release -> Completed) required meticulous attention to Next.js Server Actions. Additionally, transitioning our chat interface from standard server-polling to true Realtime WebSockets required careful React state management to ensure an instant, optimistic UI without memory leaks.

Accomplishments that we're proud of

We are incredibly proud of our API-first AI architecture. By routing our natural language parser through Featherless.ai, we managed to build an enterprise-grade "Smart-Match" algorithm that runs in milliseconds on Vercel Edge Functions without crashing. We also successfully built robust, distinct dashboards for both Clients and Pros that connect seamlessly in the middle.

What we learned

We learned the incredible power of deeply integrating Supabase Realtime with Next.js 15. We also learned how to effectively prompt and extract strict JSON payloads from open-source LLMs to power database mutations, proving that you don't need heavy, expensive proprietary models to build world-class AI features.

What's next for Guild

Our next step is implementing AI Fraud Guard—an AI listener in the chat room that flags unsafe behavior or attempts to bypass the Escrow system (e.g., "pay me outside the app"). We also plan to roll out our Premium Bidding Algorithm to help top-rated artisans boost their visibility on emergency jobs.

Built With

• featherless-ai
• llama
• next.js
• openai-sdk
• postgresql
• react
• stripe
• supabase
• tailwindcss
• typescript
• vercel