Projects

Exploring vector embeddings for document search. Comparing OpenAI, Cohere, and open source models across quality, latency, cost, and operational complexity. Built with pgvector for similarity search and Next.js for the interface.

Visual workflow builder for non technical users to automate business processes. Features include drag and drop canvas, node library with triggers, actions, conditions, delays, webhooks, and approvals. Designed with multi tenant architecture, versioning, and observability.

Research project on building reliable AI agents for business automation. Focused on correctness, safety, predictability, and cost control. Testing agent patterns including tool calling, planner executor, reflection, and retrieval assisted approaches. Demo use cases include support triage, invoice classification, and CRM updates.