From research to production.

Senior AI/ML Professional with over 5+ years in Data Science, AI, Machine Learning and MLOps across healthcare, automotive and financial services. Currently a Graduate Research Assistant at the University of Maryland, College Park, focused on 7B–70B parameter LLMs and diffusion models for biomedical applications.

Previously, as Senior AI/ML Engineer at Aya Healthcare, I architected production multi-agent LLM platforms processing 2.3M+ patient profiles with 95% clinical accuracy and built scalable RAG infrastructure serving 2,000+ concurrent providers across 15+ hospital networks.

At Bridgestone Group (Azuga Inc.), I led cloud operations, data architecture and MLOps - designing infrastructure that processes 24 billion sensor records (120 PB) with < 60s latency using Apache Spark on 24-node clusters.

Highlights include the Accident Risk Survival Model using Cox regression (C-index 0.78) on 7.8M crash records projecting $122.9M+ savings, NLP sentiment systems, a real-time streaming warehouse with 99.8% data accuracy, and a Generative AI "Chat with Data" initiative powered by LangChain & AutoGen.

Open Source · GSoC 2026 with The Linux Foundation

Selected as a Google Summer of Code 2026 contributor with The Linux Foundation, working on the Accord Project's APAP / MCP server - the reference implementation for exposing smart legal templates and agreements to AI clients via the Model Context Protocol.

The project (Idea #4: 175 hours over 12 weeks, mentored by Niall Roche and Dan Selman) covers a shared service-layer refactor that replaces the internal HTTP loop, four-tier testing (unit, integration, contract, E2E) with 90%+ coverage enforced in CI, and multi-client documentation for Claude, ChatGPT and MCP Inspector. A working proof-of-concept already ships with 53 passing tests and 98.55% statement coverage, plus five prior accepted contributions to the Accord codebase. Full case study →

Open-source contributions to other Linux Foundation projects:

• Kubeflow (CNCF) — kubeflow/docs-agent: Contributing toward production-hardening the docs-agent with CI/CD pipelines, test coverage, monitoring, RAGAS benchmarks and idempotent ingestion. Contributions to date include three filed issues (#181 content truncation discarding 60% of indexed content, #182 fragile Feast VARCHAR monkey-patch, #183 SentenceTransformer + Milvus reload compounding to ~3s overhead per search), KEP document comments on the Infrastructure and Ingestion sections, and a working end-to-end POC repository demonstrating the fixes against a Kind cluster.
• Jenkins (CDF) — resources-ai-chatbot-plugin: Designed and prototyped a context-aware, multi-agent extension to the Jenkins resources-ai-chatbot-plugin that reads live instance state (build logs, installed plugins, job configs) and routes queries through an LLM intent router to four specialised agents - Troubleshoot, Workflow, Recommend, General. Built on open-source LLMs (LangChain, FAISS + BM25 hybrid retrieval, Llama 3.x via llama.cpp, Ollama for model management), running fully self-hosted with no vendor lock-in.

Beyond the Accord Project work, my GitHub portfolio (JayDS22) spans 30+ production-grade projects including the AgentForge Multi-Agent RAG Platform (98.5% success rate, LangGraph orchestration), a Real-Time Recommendation Engine (sub-100ms latency on PySpark + Delta Lake), and a Transportation Demand Forecasting System (MAPE 2.8% with LSTM).

Additional work spans Driver Behavior Analytics using Bayesian hierarchical modeling, a Real-Time Experimentation Platform built on Thompson sampling bandits, and a Marketplace Optimization Engine (Hungarian algorithm + linear programming) at 97.2% matching efficiency.

As an Analytics Engineer at Simplilearn, I built enterprise data pipelines and advised Data Science clients at Purdue and UMass.

Published research includes "Predictive Maintenance in Automotive using Machine Learning" (accepted, IEEE Journal, Feb 2025) and "Optimizing Supply Chain using Data Science and AI" (IJAET, 2023). Earlier work at Dhirtek Business Research applied statistical modeling to liquid biopsy market growth.

I focus on building reliable, production-ready AI - from HIPAA-compliant healthcare systems to real-time fraud detection processing 500K+ daily transactions at 94.2% accuracy. The aim is always the same: scalable systems that deliver measurable business impact under real-world constraints.