A full-stack software engineer with depth in backend systems, distributed infrastructure, and applied ML.
I build production-grade systems end to end, from REST APIs and real-time messaging services to fine-tuned language models and retrieval pipelines. My work sits at the intersection of software engineering and applied ML: I care as much about how a system is deployed and scaled as I do about the model inside it. I enjoy working on projects that bring different domains together to produce an impact on the world.
Currently finishing my MS in Information Science at the University of Pittsburgh, where I also mentor Masters of Data Science capstone teams through applied data science projects. Previously a fullstack intern at TeamLease EdTech and a mobile application intern at GameOn Technologies.
Open to software engineering and ML engineering roles starting May 2026 (F-1 OPT).
Languages: Python, R, Java, C, C++, C#/.NET, JavaScript, TypeScript, SQL, HTML/CSS, PHP
ML & AI: PyTorch, TensorFlow, Scikit-learn, XGBoost, LightGBM, ONNX, Gensim, spaCy, SHAP, Plotly, NLTK, Hugging Face, Matplotlib
MLOps & data: MLflow/W&B, Airflow, Spark, Hadoop, Pandas, NumPy, SciPy, statsmodels, NetworkX
Backend & APIs: Spring Boot, Hibernate, ASP.NET MVC, Node.js/Express, Flask, FastAPI, React/Next.js, Angular, Vue.js, REST, WebSockets, OpenAPI
Infrastructure & DevOps: Docker/Kubernetes, AWS, GCP, Terraform, CI/CD, Linux, Git, CMake, GDB
Databases & tools: PostgreSQL, SQL Server, pytest, MongoDB, SQLite, Cassandra, Neo4j, Elasticsearch, Redis, BullMQ, Visual Studio, Splunk, Figma, Flutter, RaspberryPi
These are some selected projects that I've worked on recently.
- Litmus: S3-compatible object store built for correctness. AWS CLI and boto3 work unmodified. A two-tier chaos suite proves durability invariants hold across SIGKILL and injected OSErrors. 115/325 ceph/s3-tests pass; every failure documented.
- ITCH 5.0 Parser: C++ limit order book reconstruction for NASDAQ TotalView-ITCH 5.0. Lock-free snapshot handoff via
atomic<shared_ptr>, fixed-point price arithmetic throughout, 57M msg/s parse throughput. Validated against a real 7.7 GB production capture. - Botnet C2 Detection: graph topology classifier over CTU-13 (13 captures, 7 botnet families). Leave-one-family-out cross-validation with pre-registered thresholds. Donbot achieves 0.872 PR-AUC on an unseen family; the
cc_dst_onlycollapse to 0.001 is itself a documented finding. - Storm: real-time messaging platform in TypeScript. AES-256-GCM encryption, single-use JWT refresh tokens with replay detection, 112 tests against real MongoDB and Redis.