Tejas Nafde

I’m a Software Development Engineer at GeoIQ (a Lenskart company), where I built and re-architected the core backend for an AI-driven WhatsApp analytics assistant used by operations teams. I pair LLM-driven workflows with strong backend fundamentals—native BigQuery query generation, role-scoped data access, and Vertex AI (Gemini) model routing—to keep systems reliable and cost-aware.

Beyond the assistant, I’ve shipped end-to-end automation that turns conversations into operator-ready outputs: an automated weekly LLM insights pipeline using Gemini + Puppeteer PDF export, and ReviewIQ—an LLM-powered PR-review bot for Bitbucket on Gemini with two-tier review, deduped comments, and per-run cost estimation. I also built “ssg-doctor,” an autonomous RCA-and-fix agent service that polls issue trackers, reconstructs incidents via ad-hoc evals, pulls ground truth from BigQuery and Loki logs, and posts a structured evidence-chain verdict—optionally committing fixes to a review branch.

I design for operational performance and maintainability: a Postgres-backed job queue with atomic claims, orphan requeue on restart, agent account rotation with rate-limit failover, and Slack-streamed live progress. I’ve also delivered major platform improvements (replacing Airbyte refreshes with BigQuery-to-Postgres bulk-sync DAGs to cut refresh from 8–12 hours to ~30 minutes; reducing report API latency by 80% with MongoDB section caching; building an admin RBAC system; and standing up observability with OpenTelemetry + Grafana/Prometheus/Loki). Earlier, during my GeoIQ internship, I architected a real-time WebSocket + Pub/Sub messaging platform across multiple production applications, and I’m currently building additional cross-platform apps and agent tooling through my projects.