Applied AI Engineer & Researcher — building enterprise agentic systems, LLM evaluation, and model compression. Ph.D., IIT Delhi.
I turn frontier AI research into production systems: if it hallucinates, I tame it; if it's too big, I compress it.
Website · Google Scholar · LinkedIn · Email
Senior Applied AI Engineer @ KPMG — architecting enterprise-scale agentic systems.
- Shipped a globally adopted, skill-based agentic framework (FastAPI · React · Azure OpenAI) with >80% less PoC effort, 98% fewer hallucinations.
- Built an LLM-as-judge evaluation platform adopted by 50+ teams, scoring 100+ proposals concurrently with explainability.
- Responsible AI by design: PII redaction, prompt-injection guardrails, Azure AI Content Safety, while cutting inference cost up to 30%.
- TVA-prune — VIB-based structured pruning for LLaMA/Mistral: 50% compression, up to 80% inference speedup, 10–20× faster than comparable methods.
- VIB-LSTM Compression — action-recognition compression: 70× smaller, ~100× Raspberry Pi speedup vs. full LSTMs (WACV 2021).
ICML 2024 (Workshop) · WACV 2021 · PReMI 2023 — full list on Google Scholar.
Multi-agent systems · tool calling · context management · Knowledge graphs · LLM evaluation & hallucination detection · model compression (LoRA, PEFT, quantization) · hardware-aware NAS.
Open to conversations with teams that need someone fluent in both deep-learning research and production architecture.