Mehul Basu

University of Wisconsin-Madison, Madison, WI | GPA: 3.92/4.0

University of Wisconsin-Madison, Madison, WI | GPA: 3.84/4.0

• Architected a scalable full-stack roommate-matching platform (React, Supabase) with a custom matching engine to calculate real-time compatibility scores, reducing infrastructure costs by 60% through a hybrid NoSQL/SQL design.
• Managed a 4-engineer team and product roadmap while leading investor outreach and go-to-market strategy to validate market fit.
• Optimized platform performance with 24-hour client-side caching, route-level code splitting, and browser-side image compression, resulting in 68% lower latency and a 30% reduction in serverless compute costs.

• Reduced production bug backlog by 20% through triage, root-cause analysis, and targeted fixes for the Microsoft Teams app.
• Redesigned the dashboard and integrated SSO, improving daily active usage and user onboarding by 25%.

• Modernized the data upload pipeline, simplifying ingestion and enabling near-real-time analysis for large datasets.
• Proposed efficient database algorithms (Python, Pandas, MSSQL) to accelerate query times by 99.5%, from over 30 seconds down to 0.15 seconds.
• Communicated with the client to align project deliverables and identify optimization opportunities across large CSV files.

• Operate CRM systems to manage service requests from more than 800 faculty and students.
• Automated attendance reporting for 100+ classes using the Canvas LMS API and Google Apps Script.

• Developed a hybrid convolutional neural network (PyTorch) with parallel 1D-CNN and MLP branches to classify encrypted QUIC traffic with 89.5% accuracy, a 26% improvement over Random Forest baselines.
• Applied featurization combining packet-level sequences (inter-arrival times, directions, sizes) with 68 derived statistical features.
• Engineered distributed training with Distributed Data Parallel across 4 GPUs to process more than 40 million flow records.

• Engineered a weather data platform using gRPC, HDFS, and Cassandra, backed by a scalable Apache Spark data pipeline.
• Implemented data replication and LRU caching, improving uptime to 99.9% during simulated node failures.

• Leveraged OpenMP to parallelize Bellman-Ford negative-cycle detection across 19 currency node pairs.
• Achieved 3x speed-up at 4 threads and 3.6x at 8 threads vs baseline, processing 86,400 snapshots per second and simulating a theoretical 0.71% daily return on real-world forex data.

• Led a 5-person Agile Scrum team to build a full-stack app (MySQL, React, Express) with GitLab CI/CD pipelines.
• Designed ERD models, created APIs, and organized role-based views for students and professors to track requests.