I build the engine
under the product.

A multi-agent LLM platform that turns a plain-English question into governed SQL, charts and insights. I built the backend and the chart-reasoning engine.

• Multi-agent LLM→SQL over LangGraph/LangChain with real-time SSE streaming through nested ReAct agents.
• A generic runtime-connector engine unifying 8 data sources (Snowflake, Salesforce, Shopify, Stripe, HubSpot…) with opaque-cursor pagination & content-hash tamper protection.
• SQL safety & cost: read-only + SQLGlot-parse + AST minimal-change across 8 dialects; two-tier Redis→Postgres schema cache; per-org token metering.
• A chart-reasoning engine that decides which charts are meaningful (not just renderable), with lossless chart-type switching → Vega-Lite.

• Python
• FastAPI
• LangGraph
• Snowflake/BigQuery
• Redis
• Celery
• Postgres
• React

The live console ↑ runs a slice of this.

Led the team building the LLM voice-interview platform and worked across the full stack — including the real-time speech-to-speech (STT ↔ TTS) pipeline. 2026 MIT Solve ↗ global winner (1 of 6, $200K grant).

• Led the team & technical direction — planning, code reviews, system design, and staging & production (AWS EC2 + Amplify).
• Built the real-time speech-to-speech pipeline — STT ↔ TTS with voice-activity detection and client-side orchestration for sub-3s turns.
• Migrated to a fully client-side React architecture — −60% latency and lower inference cost; edge Cloudflare Worker streaming completions with the provider key off the browser.
• Interview/invite lifecycle state machine, async leveled-scoring pipeline, Firestore→Postgres migration, proctoring, bounce-detecting bulk email.

• React
• FastAPI
• PostgreSQL
• STT/TTS
• Cloudflare Worker
• Redis

As founding engineer, I built AI-Sentinel VDP — Bugbuster Labs' AI-driven platform for continuous vulnerability discovery, real-time threat detection and automated risk response. Built fullstack, and I owned the report-generation system.

• Built fullstack across AI-Sentinel VDP with a 3-person founding team laying the groundwork.
• Designed and shipped the report-generation pipeline — findings, severity, remediation, exports.

• React
• Node.js
• Report generation
• PostgreSQL

Data pipelines ingesting & enriching U.S. government procurement — delivered a U.S. Airdrop tool end-to-end in 6 weeks.

• Scheduled scrapers, dedup & normalization over SAM.gov, DLA, PartsBase, ILS (TypeScript · Playwright).
• Owned CRM backend, integrations (WhatsApp / Facebook Business / email) and on-call production hardening.

• Node.js
• TypeScript
• Playwright
• PostgreSQL