Semai.
AI support platform that gave SMBs 24/7 coverage without the brittle chatbot tooling.
Semai was built for businesses that needed to answer customers faster, qualify leads better, and stop relying on staff to manually repeat the same support work all day. The goal was not a novelty chatbot — it was a support layer that operates on real business knowledge and fits into day-to-day commercial workflows.
I designed and shipped an AI support platform with three core layers: a knowledge-grounded assistant trained on company documentation and FAQs, a delivery surface for customer conversations and lead capture, and an operational layer for maintaining content, reviewing responses, and improving the system over time.
The stack centred on FastAPI, OpenAI, vector search, and a lightweight frontend for configuration and oversight. Response coverage extended to web, WhatsApp-style flows, and internal help surfaces.
Retrieval grounded in company documentation and FAQs, with evaluation tooling to catch regressions before customers do.
Support and sales conversations across web and WhatsApp-style flows, with lead-capture handoffs into the CRM.
Review queues, content maintenance, and escalation paths so the support team stays in control of edge cases.
FastAPI + OpenAI + vector search, deployed with observability and evaluation baked in — not a weekend demo.