Delivered the flagship clinician dashboard (React + FastAPI), powering real-time insights and documentation for 2,000+ professionals.

Re-architected CosmosDB + Blob Storage pipelines, enabling sub-second retrieval on 10M+ fax records and saving $42K/year in infra costs.

Engineered an MCP-driven system of AI agents for compliance validation, verification, billing code suggestion, and OCR automation — boosting documentation completeness to 98% and cutting claim rejections by 27%.

Embedded hybrid ML + LLM workflows (LangChain + LangGraph) for retrieval, summarization, and compliance validation, improving claim FPY from 78% → 93% and scaling to 10K+ claims/day.

Achieved 92% F1-score (BLEU=0.84) on coding accuracy and 96% OCR classification accuracy, ensuring production-grade reliability.

Built modular FastAPI microservices deployed via AWS Lambda + API Gateway for session retrieval, PDF generation, and progress note validation, achieving 99% reliability at scale.

Integrated FHIR-compliant micro services to sync patient/encounter data with PCC, achieving <200ms latency and removing manual billing payload entry.

Established MLOps pipelines (Terraform + SageMaker + GitHub Actions) with drift detection and automated retraining, reducing model downtime and accelerating iteration cycles.

Designed and delivered a full-stack vitals monitoring platform built with React and FastAPI (Python), deployed on AWS EKS for high availability and scale.

Built a patient–provider recommendation engine, combining Collaborative Filtering, Gradient Boosted Decision Trees, and Deep Cross Networks (DCN v2) to model patient histories, provider specialties, and payer rules.

Integrated LLaMA 3.1 and CodeLLaMA for prompt-driven clinical alerts and automated data pipelines, enabling clinicians to view live patient data with sub-100ms updates and reducing alert latency by 30%.

Deployed embeddings in pgvector/Pinecone for fast retrieval and built ranking APIs in FastAPI with strict validation and async I/O. The pipeline scaled to 100K+ daily recommendations, delivering sub-200ms response times, improving NDCG@10 by 18%, and increasing appointment conversions by 25% in A/B tests.

Implemented a Retrieval-Augmented Generation (RAG) pipeline with FAISS + LLaMA 3.1 on AWS S3 to provide context-aware triage classification and automate clinical audits.

Engineered predictive health risk models in AWS SageMaker, leveraging Isolation Forest for anomaly detection, LightGBM for short-term risk prediction, and LSTMs for 24–48h time-series forecasting. These models processed continuous vitals data streams, reaching 92% precision in anomaly detection and 20% fewer false positives in deterioration forecasts.

Developed multimodal AI-powered clinical workflows, embedding a WebRTC-based video consultation feature with real-time audio transcription via STT. The transcripts were passed into fine-tuned GPT-4o and LLaMA 3.1 models for summarization, and structured alert generation.

As a Teaching Assistant for CS141: Intermediate Data Structures and Algorithms, I took on various responsibilities :

Enhanced student engagement by incorporating real-world examples, leading dynamic discussions to deepen understanding, and fostering critical thinking with challenging exams.

Demonstrated leadership by managing a team of eight graders and a fellow teaching assistant, promoting a collaborative and supportive educational environment.

I managed a diverse class of 150 students, maintaining high attendance and proactive involvement through regular office hours, encouraging participation, and offering targeted guidance to boost learning outcomes.

I developed clear, fair evaluation rubrics for various assessments and received exceptional feedback.

Developed FastAPI-based microservices for metadata extraction, ETL control, and pipeline auditing, cutting manual analytics intervention by 70%. Integrated MuleSoft APIs with Python services for seamless enterprise data exchange.

Deployed production-ready recommendation services handling 100K+ daily requests with sub-150ms p95 latency, validated through A/B testing on add-to-cart and bundle take rates.

Automated dynamic reporting using Pandas + Jinja2, generating configurable Excel/PDF exports for C-level stakeholders, reducing reporting turnaround from hours to minutes.

Provisioned cloud infrastructure with Terraform, automating AWS RDS, ECS, S3, and Secrets Manager with fine-grained IAM policies. Containerized services with Docker and orchestrated via ECS Fargate, achieving 99.99% uptime across production workloads.