Salesbuddy

Field sales reps live in WhatsApp. They don't switch tools, they don't log into portals, and asking them to change behavior has never worked. So the product had to be built inside the chat — not alongside it.

Two hard constraints guided every decision from day one: it has to work entirely within WhatsApp, and it cannot compromise data sovereignty for the client. Together those two things cut a lot of bad ideas early.

Store check-ins, order execution, on-ground notes, and escalation alerts — all handled through conversation. Sales managers get operational insights and P0/P1 alerts without chasing spreadsheets or waiting for end-of-day reports.

The architecture and conversation flows were fully designed before a line of code was written. Getting the system structure right first meant fewer expensive rebuilds later.

Started on n8n and Firebase for rapid validation. Current production stack is Python, LangGraph for orchestration, FastAPI, Next.js, Firestore, and Redis — with LangSmith for LLM observability and Grafana for the rest.

The move from n8n to LangGraph was about control: once the conversation flows got complex enough, we needed explicit state management and the ability to run evals on individual nodes.

Still engaged as tech consultant — helping decide what gets built, in what order, and how. The product is in production and growing. The job now is keeping the architecture ahead of the feature list.