AI/ML Research and Computer Science Engineer focused on production-grade agentic AI systems, data infrastructure, and high-performance ML platforms. I build backend services, ML pipelines, RAG systems, evaluation frameworks, and cloud-native infrastructure for environments where reliability, scale, and correctness matter.

My work spans AI model development, domain-specific fine-tuning, ranking and recommendation systems, large-scale data pipelines, LLM evaluation, and performance optimization. I have worked on systems processing 8M+ daily records, maintained 99.5% uptime across production ML infrastructure, improved inference latency from 250ms to 120ms through CUDA optimization, and built agentic AI workflows achieving 95%+ precision on critical decisions.

• 🏗️ Engineering discipline — system design, clean interfaces, tests, CI/CD, observability, reproducibility, and maintainable code.
• 🔬 AI systems — LLM memory, RAG, evaluation, retrieval, ranking, fine-tuning, long-horizon behavior, and model reliability.
• 🛠️ Production systems — pipelines, APIs, containers, cloud infrastructure, validation gates, operational diagnostics, and model serving.
• ⚡ Performance engineering — CUDA kernels, C++ inference paths, GPU memory optimization, profiling, and low-latency serving.
• 🤝 Open-source mindset — practical fixes, regression coverage, runtime tooling, SDK improvements, and documentation that helps maintainers.

Recent public work across ML infrastructure, runtime tooling, SDKs, and HPC documentation.

Open source is where I practice careful engineering in public: focused fixes, performance work, tests, and documentation that make projects easier to use and maintain.

My work sits across a few connected areas:

Jun 2026 – Present

Building production-grade agentic AI systems for regulated insurance workflows, with an emphasis on reliability, grounded reasoning, observability, and deployment-ready backend integrations.

Key Contributions

• 🧠 Agentic AI workflows — Designed multi-step reasoning, tool-use patterns, and orchestration logic for regulated insurance processes, achieving 95%+ precision on critical decisions.
• ☁️ Production deployment — Architected AWS deployment infrastructure using Amazon EKS and backend API integrations for reliable production traffic.
• 📚 Hybrid RAG systems — Grounded AI reasoning in insurance-specific knowledge sources including policy language, regulatory guidance, and underwriting criteria, improving output accuracy to 96% and reducing ungrounded responses by 38%.
• 🔎 Observability — Built distributed tracing across LLM pipelines and agentic workflows, improving error detection and issue resolution by 67%.
• 🧪 Evaluation systems — Built benchmarking suites against insurance-domain quality standards across precision, consistency, defensibility, latency, cost, and tool-use performance.
• 🔁 Self-correction loops — Designed self-reflective correction mechanisms with human-in-the-loop synchronization, reducing system failure rate by 88%.

Python FastAPI AWS Amazon EKS Agentic AI RAG LangChain LangGraph OpenTelemetry LLM Evaluation MLOps CI/CD

Jul 2025 – Present

Researching memory governance, hallucination resistance, adversarial robustness, and reliable autonomous AI systems.

Key Contributions

• 🔬 Memory governance — Designed policy-driven memory governance middleware for autonomous AI agents, achieving 83.3% accuracy versus a 78% baseline on long-context memory tasks.
• 🛡️ Robustness testing — Built evaluation infrastructure against adversarial memory-poisoning attacks, including MINJA, PoisonedRAG, and AgentPoison attack vectors.
• 📉 Token efficiency — Reduced token usage by 47% while improving hallucination resistance, transparency, and performance stability across persistent, multi-session, multi-agent LLM systems.
• 🧭 Retrieval policy — Designed intent-aware, semantically ranked retrieval heuristics and declarative policies for memory retention, retrieval, and abstraction.
• 📄 Research output — Published MemArchitect: A Policy Driven Memory Governance Layer on arXiv.

Python LangChain LangGraph RAG Pinecone LanceDB OpenAI APIs Hugging Face MLflow Evaluation Frameworks

Nov 2024 – Jun 2026

Built data reliability, cloud infrastructure, and operational tooling for distributed research workflows where correctness, access control, and repeatability matter.

Key Contributions

• 🏗️ Cloud-native services — Architected scalable Python microservices on AWS for distributed ingestion and processing across concurrent research pipelines.
• 🚦 Operational diagnostics — Developed multithreaded C++ and Python services for workflow diagnostics and failure detection, reducing incident triage time by 60%.
• 🗄️ Data modeling — Designed relational data models and integrity enforcement mechanisms in PostgreSQL, using indexing and foreign key strategies to optimize query performance by 3x.
• ✅ Reliable pipelines — Collaborated across engineering teams to optimize data pipelines for throughput, reliability, and fault tolerance, ensuring 99.9% uptime for critical workflows.
• 🧪 Research infrastructure — Built validation frameworks, measurement systems, and benchmark datasets used across 3+ ML research projects.

Python C++ SQL FastAPI AWS PostgreSQL Airflow dbt Great Expectations Docker CI/CD PyTest IAM OAuth

Dec 2023 – Jul 2024

Led content operations, analytics, forecasting, automation, and reporting workflows supporting editorial and product decisions.

Key Contributions

• 📊 Team leadership — Led a global team of 40+ members across editorial, content operations, and analytics, including hiring, performance reviews, and team development.
• 📈 SEO analytics — Owned product analytics for SEO features, driving 54x organic traffic growth through systematic keyword optimization and engagement experimentation.
• 📉 Forecasting — Designed time-series forecasting models in Python/R to predict content traffic and revenue for quarterly planning.
• 🧰 Dashboards — Built Power BI and Looker dashboards tracking content KPIs, engagement metrics, and revenue performance across 40+ team workflows.
• ⚙️ Automation — Automated SEO scoring and content optimization using Python and NLP techniques, reducing manual work by 40%.

Python SQL Pandas NumPy SciPy Power BI Looker Google Analytics D3.js Forecasting NLP

Oct 2023 – Jan 2024

Worked on domain-specific prompt design, LLM evaluation, feedback workflows, and inference-efficiency assessment.

Key Contributions

• 🧩 Prompt systems — Designed 350+ domain-specific prompts utilizing Supervised Fine-Tuning (SFT) and RLHF workflows to improve large language model reasoning and objective alignment.
• 🔁 Evaluation pipelines — Built CI/CD-integrated prompt-quality assessment workflows, enabling 25% faster iteration cycles for model improvement.
• 🧠 Structured feedback — Delivered feedback mechanisms that increased model accuracy by 12% and reduced hallucination rates by 30% across evaluated outputs.
• ⚙️ Inference analysis — Evaluated deep learning model inference efficiency across scenarios and assessed tokenization impact on model performance and computational requirements.

Prompt Engineering SFT RLHF LLM Evaluation Inference Analysis Python Prompt Quality Model Alignment

Sep 2021 – Nov 2023

Joined as a founding engineer to build and scale the AI/ML layer of an early-stage job-matching platform from data ingestion through production inference.

Key Contributions

• 🏁 Founding engineering — Designed and scaled end-to-end ML systems serving 8M+ daily records across ingestion, training data generation, model deployment, and production inference.
• 🚀 Production ML ownership — Operated 25+ production models, maintained 99.5% uptime, and served 250K+ daily inference requests at 120ms average latency.
• ⚡ CUDA optimization — Reduced inference latency from 250ms to 120ms through NVIDIA profiling and CUDA kernel optimization, while reducing GPU memory footprint by 30%.
• 🔍 Ranking and retrieval — Built and fine-tuned transformer-based retrieval models for job-candidate matching, improving search relevance by 20% through offline evaluation and online A/B testing.
• 🗄️ Warehouse optimization — Designed dimensional data models and optimized SQL query performance from 45 seconds to 3 seconds through schema and index optimization.
• 🔄 ML pipelines — Architected Airflow-orchestrated Spark and Delta Lake pipelines on Databricks, processing 8M+ daily records in Parquet format into Snowflake-backed analytics and training datasets.
• 🐳 Production services — Containerized scalable Python microservices using Docker and Kubernetes with GitHub Actions CI/CD, integrating OpenAI and LLM inference pipelines with backend APIs.

Python C++ CUDA PyTorch TensorFlow Keras scikit-learn XGBoost BERT LangChain RAG LanceDB Pinecone FastAPI Spark Databricks Snowflake Docker Kubernetes MLflow

Arizona State University · arXiv:2603.18330

Long-running agents do not just need more context. They need governed memory: what gets stored, what expires, what is allowed back into the prompt, how contradictions are resolved, and how token budgets are spent.

MemArchitect is a model-agnostic external memory governance layer for persistent LLM agents. It treats memory as a constrained, auditable resource rather than a passive transcript or naive similarity-search log.

Python LangChain LangGraph RAG Pinecone LanceDB OpenAI APIs Hugging Face Evaluation MLflow

arXiv · March 2026

Introduced a policy-driven memory governance layer for persistent LLM agents, covering memory lifecycle management, conflict resolution, privacy controls, decay, retrieval governance, hallucination resistance, and structured evaluation for reliable autonomous systems.

LLM Agents Memory Governance RAG Agentic AI Adversarial Robustness Evaluation Frameworks

IEEE Xplore · ICECONF 2025

DOI: 10.1109/ICECONF65644.2025.11379581

Co-authored a carbon-aware workload relocation framework for multi-cloud environments using reinforcement learning, forecasting, and constraint-aware optimization to balance cost, performance, energy efficiency, and emissions.

Ray RLlib PyTorch Hugging Face Transformers LSTM Carbon-Aware Scheduling Energy Modeling Pandas Python

Python · SQL · Java · C/C++ · Go · TypeScript · FastAPI · RESTful APIs · Microservices · System Design · OOP Design · Design Patterns · API Contract Design · Event-Driven Architecture · PyTest · Unit/Integration Testing

PyTorch · TensorFlow · Keras · scikit-learn · XGBoost · Transformers · BERT Fine-Tuning · Ranking Systems · Recommendation Systems · RAG · LangChain · LangGraph · LanceDB · Pinecone · OpenAI APIs · Anthropic APIs · SFT/RLHF Workflows · Prompt/Context Management · Agentic Workflow Design · LLM Evaluation · Model Alignment

Apache Airflow · Apache Spark · Databricks · Delta Lake · dbt · ETL/ELT Pipeline Design · Batch/Streaming Processing · PostgreSQL · MySQL · Snowflake · Google BigQuery · Amazon Redshift · Data Warehouse Architecture · Dimensional/Star Schema Design · 3NF Modeling · Query Optimization · Indexing Strategies · Parquet

AWS S3 · AWS Lambda · Amazon SageMaker · Amazon Redshift · EC2 · Amazon EKS · Google BigQuery · Dataflow · Composer · Dataproc · Pub/Sub · Azure Blob Storage · Azure Functions · Azure Machine Learning · Synapse Analytics · Azure Databricks · AKS · Docker · Kubernetes · GitHub Actions · CI/CD · MLflow · Model Deployment & Serving · Feature Engineering Pipelines · A/B Testing · Model Monitoring

CUDA Kernel Optimization · NVIDIA Profiling · ONNX Runtime · TensorRT / Polygraphy · Inference Latency Optimization · GPU Memory Optimization · Query Tuning · Performance Profiling · Great Expectations · Data Validation Frameworks · Automated Quality Checks · OpenTelemetry · Distributed Tracing · Structured Logging · SLO/Error Budget Tracking · Fault-Tolerant System Design · OAuth · SAML · IAM

• 📝 ICLR 2026 Reviewer — technical review experience across modern AI research, evaluation methodology, and reliable ML systems.
• 🌱 Published researcher — memory governance for autonomous AI systems and carbon-aware multi-cloud workload relocation.
• 🤝 Open-source contributor — practical fixes, performance improvements, regression coverage, and documentation across production-grade repositories.
• 🌐 Cross-functional builder — experience translating technical systems into decisions, risks, implementation plans, and measurable product outcomes.

  

I build with production constraints in mind: correctness, latency, cost, infrastructure, observability, and maintainability. I treat tests, documentation, and clear ownership as part of the product. I write code and research artifacts with the next engineer, reviewer, or maintainer in mind.