Software engineer with 5+ years building production ML systems and real-time data pipelines — most recently as a research engineer at Stony Brook's Neurobiology Lab, where I wrote Python pipelines for calcium imaging analysis targeting a Nature publication.
I'm a software engineer with 4+ years of industry experience and an MS in Data Science from Stony Brook University (GPA 3.92, May 2026).
My background spans two distinct chapters. The first was production engineering — building fintech infrastructure at CGI and later at REGART, a RegTech startup in Amsterdam whose compliance platform served EY and HSBC across European financial markets. The second was research — joining Prof. Prerana Shrestha's Neurobiology lab at Stony Brook and building a fiber photometry analysis pipeline from scratch, no prior neuroscience required, to study how fear memories form in mice.
Those two chapters have more in common than they look. Both required learning a domain fast enough to make decisions that actually mattered, then building something reliable enough that real people could depend on it daily.
I'm now looking for ML engineering and SDE roles in AI infrastructure, production ML systems, and applied AI — particularly problems where getting from research to production is the part nobody has fully solved yet.
Six production systems across AI observability, real-time infra, multi-agent LLMs, and MLOps. Not tutorials — shipped code.
Self-hosted LLM observability platform — think a production LangSmith you built yourself. Every AI agent trace streams to the dashboard within 200ms via WebSocket. Automatic LLM-as-judge scoring (0.0–1.0), Slack threshold alerts, cursor-based pagination, Prometheus metrics, and Kubernetes deployment with HPA auto-scaling 2→10 pods.
Production Kafka operations platform — real-time event pipeline management with schema registry, DLQ management, and consumer lag tracking. GraphQL subscriptions push live updates to the dashboard. Built in Go for throughput; React + D3.js for visualization.
End-to-end LLM lifecycle: QLoRA fine-tuning Phi-3-mini-4k on MedAlpaca (10,178 samples), automated eval gates with hard quality thresholds, FastAPI serving, and Prometheus + Grafana production monitoring. Fine-tuned model is live on HuggingFace Hub.
3-agent pipeline that generates investment research memos from live SEC EDGAR filings. Agent 1 fetches data, Agent 2 writes the memo, Agent 3 (Critic) fact-checks and flags hallucinations before output. Groq Llama 3.3 70B. Type a ticker, get a PDF memo in 30 seconds.
Production RAG agent built for Shrestha Lab (SUNY Stony Brook). Routes questions across four retrieval strategies: paper RAG, code RAG, knowledge graph (NetworkX), and live PubMed search. CrossEncoder reranking. RAGAS faithfulness 0.9–1.0.
End-to-end stock price forecasting app using Random Forest regression on Yahoo Finance data. Computes 20/50/200-day moving averages, trend identification, and volume indicators. Interactive Streamlit dashboard with next-day price predictions and technical overlays.
Production systems, research pipelines, and regulatory tech — across three countries.
Not a buzzword dump — these are tools I've shipped with.
Open to opportunities
ML Engineering · SDE · AI Infrastructure · Applied AI