From building web crawlers serving 80 newspapers, to designing Big Data architectures at American Express, to leading Deep Learning teams deploying on edge devices — to now shaping enterprise GenAI platforms at Cigna. Two decades of evolution, one constant: making machines understand.
Each era didn't replace the last — it built on it. The foundation engineer became the data architect, who became the deep learning lead, who became the GenAI architect.
Senior Software Engineer @ Eterno Infotech
Built DailyHunt's backend — web crawlers aggregating news from 80 publishers across rich, XML, HTML, JSON feeds. Learned to think at scale when millions of mobile users depended on your code.
Module Lead @ Impetus (American Express)
Architected SERT — the Speed Engagement and Relevance Tool for Amex open card merchants. Processed terabytes of transactional data. Discovered that data, not code, is the real product.
DS Lead @ Inkers → Practice Lead @ Trigyn
Led CV teams building face recognition, video analytics across 1000s of streams, edge deployment on Jetson Nano. Made machines see.
NLP Architect @ Hexad (VW) → ML Advisor @ Cigna
Designing enterprise GenAI platforms — RAG pipelines, LLM guardrails, document intelligence. Making machines understand and reason.
Enterprise GenAI platforms, RAG with VectorDB/FAISS, prompt engineering (CoT, ToT, few-shot), LLM guardrails, PII safety, evaluation gates, vLLM serving
Object detection, face recognition, medical imaging (MRI/X-Ray), video analytics, model pruning, quantization, edge deployment on Jetson
LoRA/QLoRA fine-tuning, RLHF alignment, knowledge distillation, DDP/FSDP distributed training, ZeRO, FlashAttention, GPU performance at scale
Team building, architecture standards, cost/latency SLOs, security-by-design, experiment tracking, release discipline, stakeholder management
Enterprise systems, research prototypes, and everything in between
Designed the architecture for Cigna's enterprise GenAI pipeline — prompt + few-shot retrieval from custom fine-tuned models (embed+CNN) for complex table extraction. Built with FAISS, LLM guardrails, PII/data controls, evaluation gates.
Built PDF parsing using CV-based layout understanding with semantic merging + entity logic. Improved throughput 2x and extraction accuracy by +37%. Also delivered QAM for hardware knowledge using FLAN-T5 and VectorDB.
Led end-to-end platform for real-time analytics across 1000s of video streams and edge devices (Jetson Nano/TX2). Managed multi-skill team spanning embedded, backend, dashboard, and data science.
BITS Pilani, India
Deep Neural Networks, Deep Reinforcement Learning, NLP, Statistics, Information Retrieval
VTU, India — First Class
Operating Systems, Computer Networks, Parallel Programming, Algorithms & Data Structures
Vehicle Detection & Tracking, Traffic Sign Classification
PoS Tagging (HMM), Machine Translation (DNN), Speech Recognition
Generative AI with LLM, Neural Networks & DL, CNNs, AI for Medical Prognosis, Hyperparameter Tuning, and more
Sun Certified Java Programmer, Sun Certified Web Component Developer
How agentic patterns — ReAct, Plan-and-Execute, Reflexion — are transforming AI from question-answering to goal-directed problem solving.
ReadWhy fine-tuning a 3B model often outperforms prompting a 70B one — at 25x lower cost and 10x lower latency.
ReadOpen to architecture consulting, advisory roles, and research collaborations