I like building the systems around AI models that make them reliable, observable, and useful in production.

• 🎙️ Real-time Voice AI — STT, LLM, TTS, turn-taking & interruption handling
• 🤖 LLM Systems — agents, context, retrieval, evaluation & structured execution
• 🔎 Retrieval & RAG — hybrid search, reranking, provenance & grounded generation
• ⚙️ Backend Systems — APIs, distributed workloads, data pipelines & infrastructure
• 🧩 Developer Infrastructure — API contracts, ASTs, compilers & validation systems

Building production AI infrastructure for real-time voice and omnichannel communication.

• Architecting an omnichannel Voice AI platform with pluggable telephony and SIP transports
• Designing context architecture that separates task policy, conversation state, user-response signals and model output
• Engineering real-time turn-taking and interruption handling across VAD, STT, LLM and TTS
• Building modular AI infrastructure with provider abstractions, local/hosted services and cost instrumentation
• Working on production observability, reliability and end-to-end latency

• Led backend and AI development for a multi-tenant NL2SQL platform capable of querying arbitrary ERP databases
• Designed an ontology-inspired semantic layer for schema entities, relational roles and consistency constraints
• Replaced embedding-only schema mapping with deterministic scoring, confidence gates and a semantic DSL

RAG system over SEC EDGAR filings designed around grounded answers and retrieval provenance.

RAG pgvector BM25 Reranking NeMo Guardrails SSE

• Hybrid retrieval over 1,000+ document chunks
• Citation-backed answers with retrieval provenance
• Guardrails against adversarial and off-topic queries
• Real-time response streaming

Developer tooling that discovers API contracts directly from source code.

TypeScript AST GraphQL OpenAPI 3.1

• AST-based API contract discovery
• GraphQL introspection
• Framework-agnostic Contract IR
• Compiler pipeline for OpenAPI generation
• Automated validation, negative tests and breaking-change detection

Graph-based ML system for detecting misinformation using multiple signals.

PyTorch GATv2 Knowledge Graphs NLP

• 23K articles
• 100K+ graph edges
• Temporal, content, source-credibility and echo-chamber signals
• 94.80% F1 on FakeNewsNet

🥇 1st Place — Dev With AI Hackathon, 2024

🥈 1st Runner-Up — Codestarts CodeBounty DSA Hackathon

🎓 B.Tech, Computer Science & Engineering (Data Science)
Dwarkadas J. Sanghvi College of Engineering · 2026 · 8.32/10

Languages

Python C++ Java TypeScript JavaScript

AI / ML

PyTorch TensorFlow LLMs RAG NLP Knowledge Graphs GATv2 Semantic Search

AI Systems

Voice AI STT / LLM / TTS Context Management Retrieval Evaluation Observability

Backend & Infrastructure

FastAPI Node.js PostgreSQL MongoDB Redis pgvector

Cloud & DevOps

AWS Docker Airflow Terraform CI/CD

• Distributed systems & system design
• AI infrastructure and scalable workloads
• Agent reliability and evaluation
• Event-driven architectures
• Context engineering & memory
• Real-time AI systems