● Built real-time in-app chat SDK for rider-driver messaging in Python: idempotent delivery via UUID keys, rate limiting, auto-escalation to support staff, deployed on Railway. ● Built automated monitoring tool covering 10 production services: WhatsApp/email alerts on detection, Twilio call escalation as final escalation. ● Built post-ride feedback pipeline via Gupshup WhatsApp Business API with non-blocking S3 uploads and UUID filenames, used by support agents across 6 districts. ● Diagnosed third-party Maps API firing every 10 seconds per driver device, eliminating ₹12,000/day in API overage. ● Built offline-first GPS write resilience after submarine cable outage: on-device queuing with ride ID and sequence number, idempotent server upserts, zero rides lost to network drops. ● Built self-hosted OSRM map-matching pipeline (Docker) reconstructing exact driver routes from GPS traces, correcting 5–10% GPS drift and reducing fare disputes. ● Owned full production stack: Linux, Apache, SSL, DNS, PM2, Git-push deploys over SSH, zero deployment failures. ● Directed 6-person outsourced engineering team: wrote feature specs, coordinated fixes from log and DB analysis.

● Building multi-agent system for US federal contract intelligence from a $700B+ annual pool. ● Six agents (Scout, Scorer, ProposalAI, Compliance Monitor, WinLoss, Pre-RFP Signals) run on a shared context layer. ● ProposalAI uses RAG over past proposals and self-corrects any section scoring below 7/10 before finalising.

● Built vulntriage, an open-source CVE prioritisation CLI published on PyPI. ● Pulls NVD, CISA KEV, and EPSS threat intel, feeds enriched data to an LLM, and ranks by real reachability. ● Supports Claude, GPT, Gemini, and Ollama; works offline with cached intel.

● Built Python/Playwright tool for large-scale recipe image dataset collection. ● Multi-strategy URL discovery handles inconsistent site structures. ● Rate-limited parallel workers with checkpointing recover from failures and resume mid-run.