InfoReach
A serverless AI document sales platform: an embeddable RAG chatbot grounded in each client's own docs.
- Year:
- 2025
- Role:
- Full-Stack Software Engineer
- React
- TypeScript
- OpenAI API
- RAG
- AWS Lambda
- DynamoDB
- AWS S3
- AWS AppSync
- CloudFront
- AWS ACM
- AWS SES
- Zod
oVice's AI document sales platform, serverless on AWS — clients upload their product PDFs and policies, and an embeddable chatbot (one script tag) answers customer questions grounded in that content, recommends source links, hands off to live agents, and tracks the resulting leads.
Problem
Japanese companies wanted to offer customers self-serve support and answer product questions without building custom infrastructure per client, plus a way to see the leads that support conversations generated.
Approach
Built the embeddable chatbot end to end: RAG on the OpenAI Responses API with answers streamed over AppSync pub/sub, recommended source links and live-agent hand-offs, plus an isolated test-chat mode. Owned the knowledge-ingestion backend — the DynamoDB model and upload states, and a Lambda that moves PDFs up to 100 MB from S3 into OpenAI vector stores, streaming progress over subscriptions to get around AppSync's 30-second limit, with a paginated, fail-safe check so transient errors can't wipe an index. Delivered self-service custom domains (DNS-validated ACM certificates, CloudFront multi-tenant routing, a guided four-state setup UI), seat licensing and credit-gated lead details with role-based authorization and Zod validation, and cut the dashboard's WebSocket subscriptions from N to 1 while updating query caches in place instead of refetching.
Outcome
Clients serving 4,000+ enterprise customers upload their own documents and get an AI chatbot that answers from that content on their own domain, with leads tracked in a faster dashboard.