Full-Stack Software Engineer / ML Engineer
I build full-stack products, backend platforms, and ML systems that survive real users.
My work spans React/TypeScript applications, FastAPI and Node.js services, cloud infrastructure, RAG assistants, computer vision, and production ML pipelines.
Current focus
Engineering-first ML and product work
Building systems where product behavior, backend reliability, model quality, and developer experience all matter at the same time.
Live
Full-stack agentic product
Teserax graph-based research agent
16%
Ticket deflection
CVS self-service RAG assistant
12.64M
Synchronized multiplayer frames
Solaris world-model research dataset
96.57%
DistilBERT accuracy
Cloud NLP content moderation API
Selected technical wins
The strongest signal is shipped work under real constraints.
These are representative examples of ownership across full-stack products, backend/platform systems, and production ML workflows.
Full-Stack Product Engineering
Built and shipped a live agentic research product
Production SaaS
Designed Teserax as a graph-based research agent with React, TypeScript, FastAPI, typed API contracts, live-source RAG, async orchestration, and failure recovery.
Backend + Platform Engineering
Built production internal systems at CVS Health
16% ticket deflection
Built a self-service RAG assistant over ticket history and internal docs, cut resolution time by 20%, and owned reliability for customer-facing platform systems.
ML Engineering
Built deployable ML systems, not just notebooks
96.57% NLP accuracy
Built model-serving APIs, medical imaging pipelines, language-model experiments, and world-model data/evaluation systems with reproducible training and deployment paths.
Flagship work
Projects that best represent technical depth and execution range.
A mix of shipped enterprise systems, agentic product work, and research pipelines. Each project includes architecture, decisions, and results.
SaaS Product
Teserax - Graph-Based Research Agent
Solo-built production TypeScript SaaS research agent that turns linear LLM chat into a graph-based exploration space with branching, synthesis, tool calling, RAG over live web sources, async orchestration, typed API contracts, and graceful failure recovery.
Synchronized Frames
Solaris — Multiplayer Video World Model in Minecraft
First multiplayer video world model generating consistent first-person observations for two players simultaneously. NeurIPS 2026 submission. SolarisEngine data collection system captures 12.64M synchronized multiplayer frames. Checkpointed Self Forcing enables long-horizon training at reduced memory cost with 36% FID improvement. Multi-agent evaluation framework using VLM-as-Judge methodology. arXiv:2602.22208, NYU.
Accuracy
Cloud NLP Classification on GCP
Production-ready multi-model text classification service with zero-downtime model switching, deployed on GCP Compute Engine. DistilBERT trained to 96.57% accuracy on a 24,783-sample dataset; 326+ test suite at 100% pass rate.
Accuracy
SliceWise — MRI Brain Tumor Detection & Segmentation
Multimodal MRI classification and segmentation model trained on BraTS dataset with shared encoder, achieving 91.3% accuracy and 97.1% sensitivity.
Capabilities
Broad tool coverage matters less than where the tools have already been proven.
Full-Stack Product Engineering
Built user-facing and internal products with React, TypeScript, FastAPI, Node.js, typed contracts, cloud deployment, and operational ownership.
ML Engineering
Built model-backed systems across NLP, computer vision, RAG, VLM evaluation, and multimodal pipelines with reproducible training and inference workflows.
Backend + Platform Systems
Built internal developer tooling, observability pipelines, storage, databases, cloud deployment workflows, and support systems for production teams.
Affiliated with
NeurIPS Next step
If you need someone who can own both the model and the system around it, start here.
I am most useful on teams building full-stack products, ML-backed applications, backend platforms, and agentic systems that need to survive contact with real users.