I build production data systems that process real transactions at scale and solve measurable business problems.
6+ years building enterprise data systems at HCL Technologies, where I owned ETL pipelines processing 100K+ daily transactions across SAP, Informatica MDM, and AWS. Refactored legacy SQL that cut cycle time by 45%. Built an ML failure prediction model that caught 85% of infrastructure failures before they happened (HCL Innovation Box 2023).
Before that, I was at Uttarakhand Aaj, a digital news platform operating across 13 states, where I built the analytical case that got leadership to approve a full infrastructure migration from monolithic to modular architecture.
MS in Data Science, University at Buffalo (Dec 2025). Currently seeking Data Engineer, Data Analyst, Data Scientist, and ML Engineer roles.
I don't build tutorial projects. Every repo below solves a real problem for a real user or client.
|
The Problem: A manufacturing client (Nissha Medical Technologies) produces ~30M casino tickets/day with a 4% defect rate in Q-Block alignment. Manual inspection cannot scale. What I Built: Real-time computer vision pipeline using YOLOv8 + CUDA + TensorRT with 4 deterministic validation gates. 100% accuracy on 2,177 production images. 45% latency reduction (117ms to 62ms). My Role: Team Lead. Designed the architecture, built the inference engine, defined production acceptance criteria with client engineers.
|
The Problem: Over 80% of US medical bills contain errors. Patients overpay by thousands because they cannot interpret CPT/ICD codes or generate dispute letters. What I Built: AI-powered billing advocate that ingests medical bills via OCR, detects billing errors using Google Gemini, and auto-generates legally structured dispute letters. Reduced manual reconciliation by ~70%. Recognition: Winner, AI For Good Hackathon, University at Buffalo.
|
|
The Problem: Retail traders juggle fragmented tools for charts, news, and backtesting. No single platform connects real-time data, technical analysis, and strategy validation. What I Built: Full-stack trading intelligence platform with ML-driven adaptive strategies (30-min learning cycles), sentiment analysis across 9 global exchanges, and a backtesting engine generating 1,000+ simulated trades. Containerized with Docker, deployed on Render.
|
The Problem: In production, fraud labels are rare or delayed. Financial institutions need detection systems that work without labeled training data. What I Built: Two unsupervised anomaly detection models (Isolation Forest + Autoencoder) on 284K+ real transactions. Autoencoder catches 84.6% of fraud; Isolation Forest minimizes false positives. Documented the precision-recall tradeoff as a business decision, not just a technical one.
|
| Daily Transactions Processed at HCL |
Processing Time Reduction |
Failure Prediction Accuracy |
Units/Day Throughput |
Defect Detection Accuracy |
Manual Work Reduced |
-7B1FA2?style=flat-square&logoColor=white){kind=icon}
-FF6F00?style=flat-square&logoColor=white){kind=icon}