Data Engineering · Applied ML · Search & Ranking · Backend Systems
M.S. Computer Engineering, NYU Tandon — May 2026 · 4.0 GPA
View Portfolio · LinkedIn · Repositories · Résumé · Email
- Best aligned with early-career Data Engineering, Applied Data Science, Search & Ranking, ML/Data Platform, and backend data systems roles.
- Current research: Graduate Research Assistant at NYU Chunara Lab since January 2026, building reproducible news-data ingestion and structured LLM-labeling workflows.
- Most recent industry experience: Data Science Intern — Emerging Technology at FOX Corporation / FOX Tech, February–April 2026.
- Core differentiator: connecting data pipelines, retrieval and ranking, model evaluation, typed APIs, testing, and deployment into measurable end-to-end systems.
| 4.0 / 4.0 NYU Tandon M.S. GPA |
12,168 profiles hybrid retrieval at <350 ms p95 |
| ~13M records historical MTA analysis |
92.7% fewer bytes scanned in SignalLake benchmark |
NYU Chunara Lab — Graduate Research Assistant · Jan 2026–present
Building reproducible pipelines across 10 U.S. newspapers that process 7,000+ articles weekly, plus validated LLM-labeling and external research datasets. Experience details
FOX Corporation / FOX Tech — Data Science Intern — Emerging Technology · Feb–Apr 2026
Built a 9-stage Databricks pipeline and hybrid semantic clustering/ranking workflow for editorial content; validated duplicate-safe runs across 14 dates in 32–42 seconds. Experience details
Global Futures Group — Software Engineer Intern, AI/ML & Data Infrastructure · Sep–Dec 2025
Built hybrid FAISS and BM25 retrieval, ranking, FastAPI, and PostgreSQL infrastructure for 12,168 expert profiles with p95 search latency below 350 ms. Case study
NYU DICE Lab — Graduate Research Assistant · May–Dec 2025
Implemented LoRA/PEFT experiments across DistilGPT-2 and Pythia models to measure Dynamic Tanh versus LayerNorm quality and inference trade-offs. Research project
SignalLake — Operational log analytics
Local-first ingestion and analytics platform that preserves raw JSONL events, transforms them into Hive-partitioned Parquet, and queries operational metrics directly with DuckDB.
FastAPI · DuckDB · Parquet · Pydantic · Docker
Evidence: benchmarked 1M events across 192 files; partition pruning reduced data scanned from 86.7 MB to 6.3 MB. Validated with 21 tests across four pytest files and GitHub Actions. Case study
ExpertMatchAI — Hybrid expert search
Internship project combining FAISS vector retrieval, BM25 lexical search, structured geo filters, tunable ranking weights, and fallback logic behind FastAPI and PostgreSQL services.
FAISS · BM25 · FastAPI · PostgreSQL · Next.js
Evidence: indexed 12,168 profiles; average response time below 200 ms, p95 below 350 ms, and full index rebuilds below 60 seconds. Validated with pytest, Vitest, and Playwright. Proprietary source code is not public.
NYC Subway Foot-Traffic Forecasting — Streaming ML
Combined analysis and model development over approximately 13M historical MTA records with a separate live self-hosted pipeline using simulated turnstile events.
Kafka · Spark Structured Streaming · MongoDB · Random Forest · Docker
Evidence: regression models reached approximately 2,700 RMSE and below 4.8% MAE; a separate traffic-level classifier reached 93.36% accuracy. Case study
Colorectal Cancer Survival Prediction — Reproducible MLOps
Public-healthcare-data MLOps prototype covering feature selection, Gradient Boosting training, experiment tracking, Kubeflow orchestration, and Flask model serving; not a clinically validated system.
scikit-learn · MLflow · DAGsHub · Kubeflow · Flask
Evidence: processed 167,497 records, reduced 28 inputs to five through chi-square selection, and reached 92.9% accuracy with 0.89 ROC-AUC. Case study
| Role family | Strongest public evidence |
|---|---|
| Data Engineering | FOX Databricks pipeline, current NYU ingestion, SignalLake, Kafka/Spark streaming |
| Applied Data Science / ML | FOX clustering and ranking, NYU research, forecasting, reproducible MLOps |
| Search & Ranking | Global Futures / ExpertMatchAI hybrid retrieval, geo filtering, tunable ranking, fallbacks |
| Backend Data Systems | FastAPI, PostgreSQL, DuckDB, REST APIs, Docker, CI, structured logging, automated tests |
- Data & Distributed Systems: Python, SQL, PySpark, Spark Structured Streaming, Kafka, Databricks, Delta Lake, Parquet, DuckDB
- ML, Retrieval & Evaluation: scikit-learn, PyTorch, Hugging Face, FAISS, BM25, embeddings, MLflow, relevance and model evaluation
- Backend & Databases: FastAPI, Flask, PostgreSQL, MongoDB, Prisma, Next.js, TypeScript, REST APIs
- Cloud, Testing & Delivery: Docker, Kubernetes, Kubeflow, GitHub Actions, Vercel, Railway, pytest, Vitest, Playwright
Start with the simplest correct system, define measurable behavior, and make failures visible before adding complexity.
- Evaluate retrieval and ML systems with explicit quality, latency, and failure criteria.
- Make data and model workflows reproducible through versioned artifacts, rerunnable pipelines, tests, and CI.
- Document architectural trade-offs so another engineer can understand both the result and its limits.
I am currently building reproducible text-data and LLM-labeling systems at NYU Chunara Lab and am open to early-career roles in Data Engineering, Applied Data Science, Search & Ranking, ML/Data Platforms, and backend systems for data and AI.
View the portfolio · Explore repositories · Connect on LinkedIn · Email me