"An Independent Researcher & Production-Grade Builder."
I am an AI/ML Engineer who builds intelligent systems with measurable impact. My work sits at the intersection of complex data, advanced algorithms, and unforgiving real-world constraints—transforming theoretical problems into highly scalable, production-ready solutions.
I specialize in designing end-to-end machine learning pipelines from the ground up: encompassing rigorous data ingestion, feature engineering, model deployment, continuous monitoring, and automated improvement. I bring a deep foundation in statistical learning, deep learning, and mathematical optimization.
Crucially, I pair this theoretical rigor with battle-tested experience in delivering infrastructure that survives under extreme operational pressure—a discipline I honed extensively during my tenure in Production Management at J.P. Morgan Chase (a non-AI role that taught me how to operate flawless mission-critical systems). I care deeply about the engineering craft: reproducibility, model interpretability, and robust evaluation are never afterthoughts; they are my core principles. I understand fundamentally that the greatest algorithms only matter if they integrate flawlessly with existing systems, empower teams, and drive concrete business objectives.
In short: I don't just train models; I engineer intelligence. Bridging the gap between academic rigor and production reliability is my specialty.
When I'm not optimizing data pipelines or researching novel architectures, you'll find me singing 🎤, travelling to new corners of the globe 🌍, or dissecting anime lore. I'm also an avid reader 📚 and a passionate chess player ♟️—always exploring new worlds, whether in code or in reality.
📍 India • 🎓 B.E. Information Technology (9.59/10) • 🏢 Ex-J.P. Morgan Chase (Production Management)
|
A production-grade, 14-stage automated pipeline I authored for high-fidelity structured synthesis. FLSS fundamentally solves single-model hallucinations by forcing an adversarial consensus mechanism. It goes far beyond simply chaining API calls, instead orchestrating a highly complex reasoning engine:
This complete rewrite has systematically dropped hallucination rates down to 0.31% over 1,270 complex data objects. |
Languages
AI & Machine Learning
Web & Backend
Cloud, DevOps & Tools
|
|
Production-grade 14-stage pipeline for structured content generation with adaptive conditional compute. Processes 1,270 aerospace engineering queries with 93.5% precision via Hybrid RAG and multi-model consensus.
|
Downstream application of the FLSS consensus pipeline. Automatically induced a curriculum-aligned ontology with 37,970 nodes and 55,153 edges directly from synthesized question metadata. Achieved 94.95% connectivity without manual curation, enabling bidirectional navigation (Questions ↔ Concepts) and exposing hidden semantic relationships (prerequisites, common mistakes).
|
|
DeepFake detection system using Inception-ResNet-v2 with Swish/Mish activation optimization. Metadata extraction for forensic video origin verification. 5% accuracy improvement over ReLU baselines.
|
Conversational AI assistant for hotel booking built with the Rasa framework. Natural language understanding with intent classification and entity extraction for hospitality domain.
|
