Résumé

• Owned end-to-end development of two applications for wire-bonding operations at Micron's Sanand semiconductor manufacturing site, spanning system design, model development, APIs, dashboards, testing, monitoring, and technical handoff.
• Built and deployed a ResNet-50 defect-classification system that achieved 88% macro-F1 on a held-out test set. Developed its FastAPI backend and React dashboards with PostgreSQL, Redis, Microsoft Entra ID authentication, RBAC, rate limiting, model versioning, and prediction logging, reducing defect review and reject-code assignment time from 3-5 minutes to about 30 seconds.
• Built and deployed an Airflow ETL and alerting pipeline for 400 wire bonders to detect gold-wire consumption anomalies associated with machine wear, misconfiguration, or theft. The system combined hysteresis control limits, CUSUM excursion, EWMA max-slope, and intra-lot variance checks, achieved a false-positive rate below 1% across 180 days of historical data, and reduced per-machine analysis time from 15 minutes to 1 minute.
• Containerized and deployed services on OpenShift/Kubernetes, added unit, integration, and end-to-end tests and Prometheus/Grafana monitoring, and wrote architecture documents, setup guides, and troubleshooting procedures to support deployment, handoff, and maintenance.

Technologies: Python, PyTorch, FastAPI, PostgreSQL, Redis, React, TypeScript, DVC, Apache Airflow, Docker, Prometheus, Grafana, Microsoft Entra ID, OpenShift, Kubernetes

• Developed Solution Tester, a no-code UI regression-testing framework for TraceLink's OPUS platform that allowed teams to validate configurable customer pages through reusable BDD scenarios.
• Designed the Playwright-based test engine and custom parser that converted human-readable BDD specifications into executable browser tests. Integrated more than 20 critical platform paths into the nightly regression suite.
• Built a Jira release-notes generator that used JQL to collect feature packs, epics, and stories added since the previous release and compile them into structured Markdown release notes.

Technologies: JavaScript, Playwright, BDD, Jira, JQL

• Developed ART, an adaptive resampling method that updates the training distribution using class-wise validation macro-F1 scores and empirical class priors.
• Built a reproducible experiment pipeline that compared ART with 11 class-imbalance methods across five tabular, image, and text datasets using 20 fixed-seed runs per method-dataset pair. ART improved macro-F1 by 2.64 percentage points on average across the tabular datasets compared with training on the original imbalanced data.
• Validated the results using Wilcoxon signed-rank tests at p < 0.05 and released ART as a reusable PyTorch package on PyPI.

Technologies: Python, PyTorch, scikit-learn, NumPy, pandas, Optuna, pytest

• Built a survival-analysis pipeline using TCGA glioblastoma clinical and somatic-mutation data to study mutation patterns associated with patient outcomes.
• Combined PyTorch autoencoders, K-means clustering, chi-square tests, KEGG pathway enrichment, Cox proportional-hazards modeling, Kaplan-Meier analysis, and log-rank tests.
• Identified six candidate survival-associated genes and built an inference tool that generates individualized survival-probability curves from their mutation status.

Technologies: Python, PyTorch, scikit-learn, GSEApy, SciPy, pandas, NumPy, Matplotlib, Seaborn

• Built a UCI-compatible C++ chess engine that earned independent Elo ratings of 3573 in CCRL Blitz and 3556 in CCRL 40/15 and won the Young Talents CEDR tournament with a score of 34.5/36.
• Implemented the core engine using bitboards, legal move generation, Zobrist hashing, transposition tables, iterative-deepening alpha-beta search, aspiration windows, move ordering, and advanced pruning methods.
• Trained and integrated a sparse 768-input NNUE model that improved playing strength by about 500 Elo. Optimized inference with AVX2/AVX-512 SIMD, added Lazy SMP multithreading, and evaluated changes through SPRT testing.

Technologies: C++20, Python, NNUE, SIMD, AVX2, AVX-512, Make