DocsChat

Inspiration

I work in a technical customer-facing role at a database company, and I spend too much of my time answering customer questions with information that is readily available in the documentation. The vast majority of people don't want to read - they just want quick answers. I'm building DocsChat to provide quick answers to technical questions based on the Neo4j documentation.

What it does

Allows a user to visit the website, ask technical questions related to Neo4j, and receive accurate ChatGPT-style answers in response.

How we built it

From an architecture perspective, it's a Go web app that takes the Neo4j documentation files, uses OpenAI's 'text-embedding-ada-002' model to create embeddings that are subsequently stored in Pinecone for indexing. The relevant context gets injected along with the user prompt to the OpenAI's LLM and the answers are streamed back to the UI. The app is running in a mostly serverless fashion on GCP.

Challenges we ran into

I'm not a professional software developer, I've only built a handful of web apps before. The whole process was a challenge, and required learning multiple new concepts in short order.

Accomplishments that we're proud of

This is the first time I've tried to build anything AI-related, and I'm walking away from the experience feeling motivated to continue building.

What we learned

Copilot in VScode is truly awesome.

What's next for DocsChat

Using it in my day job, and sharing it with my colleagues. Also, potentially handing the project over to our DevRel engineers to further build it out, and eventually turn it into a proper customer-facing application.

Built With

• gcp
• go
• langchain
• openai
• pinecone