At IIT (BHU), I implemented Google’s Wide & Deep recommender model in TensorFlow, using a custom gradient loop and feature crosses across 1M+ records. The model reduced baseline RMSE by 18.7% and achieved 0.72 MAE.
I also built a C++ multithreaded task execution engine with priority-based scheduling, mutexes, and condition variables. It benchmarked 100K task dispatches in about 385 ms across 10 worker threads.
For IIT BHU’s Convocation Portal, I built a Next.js system for hostel booking and automated allocation, then implemented secure access and a payment-to-pass workflow. The workflow issued 700+ passes and reduced manual effort by 80%; I’ve also led SDE work on a club website, ERP system, and AI-driven analytics solutions.

