Ketul Kishorbhai Chhaya
AI-focused backend engineer. Architected BizChat, an LLM platform serving 220+ Baltimore entrepreneurs via the Mayor's Office, with routing across Bedrock/OpenAI/Anthropic and an open-source observability package.
Experience
Software Engineer – Platform & AI
BizChat- Shipped a multi provider model router adopted by 64% of active users, enabling mid-session switching across OpenAI, Claude, Mistral, and Llama with zero context loss via Firebase persisted session state.
- Engineered a TypeScript/Firestore pipeline aggregating 100K+ telemetry events into human-AI collaboration metrics, surfaced via a React dashboard for a CHI publication.
- Reduced monthly LLM costs by migrating a scraping workload that drove 40% of token spend, identified via a self built open-source TypeScript proxy (tokenwise-tracker).
- Built a streaming AI edit pipeline in TypeScript, Next.js, and AWS Bedrock, where the LLM emits NDJSON ops against a custom ProseMirror section schema, replacing a 700-line fuzzy matcher.
- Shipped inline accept/reject diffs with per-edit outcome logging, creating a human-in-the-loop funnel measuring AI suggestion acceptance.
- Automated an end-to-end Playwright test suite covering the full user journey across Chromium, Firefox, and WebKit, with Slack alerts and report links that caught regressions before production.
Software Engineer
Aavenir- Cut contract migration from two weeks to five days with a Python ETL pipeline on ServiceNow workers that moved 8,000+ contracts into ContractFlow using AI metadata extraction.
- Prevented 6+ concurrent redlining collisions on shared contracts with a JavaScript webhook based locking and presence system between the Office.js Word add-in and ContractFlow portal.
- Designed a provisioning engine that aggregated config across existing client environments into a reusable baseline, auto setting up new ContractFlow clients and saving 100+ engineering hours across 4 deployments.
- Scaled a NestJS approval microservice to route contracts through parallel or sequential approver chains by value, cutting approval turnaround 50% over the prior manual process.
- Implemented an AngularJS admin console to configure approvals, triggers, delegations, and audit timelines, cutting manual workflow effort 35%.
Full Stack Engineer Intern
ClosestCloset- Optimized MongoDB ranking API from 5s to 800ms by offloading a global aggregation to a Node.js cron job that precomputed and cached scores every 2 hours.
- Launched a real-time Node.js chat service on Azure Web PubSub with a React client linking buyers and sellers, with email fallback for offline users, cutting order-status support tickets by 40%.
- Built an internal CMS in React, Express, and MongoDB that let marketing self-publish campaign pages and edit site content, removing roughly 12 engineering tickets per week.
Projects
Building LangGraph and Temporal pipeline that crawls sites into pgvector, probes Claude/GPT/Gemini in parallel on buyer intent questions, and scores brand visibility via cosine similarity gaps with sourced fixes.
Designed a fault-tolerant, event-driven payment pipeline (FastAPI, Kafka) achieving exactly-once processing guarantees via Redis SETNX idempotency keys, with exponential backoff retries and a dead-letter queue to isolate failures.
Built a TypeScript MCP server that parses a codebase into an import dependency graph, helping AI agents catch circular imports that break builds and trace what a file change affects across 1,448 files in 1.1s.
Designed a TypeScript production-scheduling engine that reflows orders around delays and maintenance windows via Kahn's topological sort and greedy shift-aware placement, covered by 16 Vitest cases.