Career Profile
Focused on the reliable deployment and interpretable reasoning of LLMs in healthcare, with hands-on expertise in prompt engineering, RAG optimization, AI safety and risk control, and context engineering. Adept at structuring clinical reasoning into executable AI inference workflows to build secure, compliant, and high-accuracy digital clinician systems. Combines full-stack engineering capabilities with strong product sense, committed to addressing high-impact clinical challenges through trustworthy AI.
Experience
- Digital Doctor Enablement: Spearheaded the development of a “Digital Doctor Agent” by digitizing real-world physician experience from Electronic Health Records (EHR) into computable assets.
- Experience Extraction System: Designed a 3-layer “Visit-Trajectory-Knowledge” experience card architecture to structure clinical logic, serving as the cognitive core for the agent’s decision-making.
- Safety & Evidence Control: Implemented “Controlled RAG” with strict evidence citation, ensuring the agent’s medical advice is traceable, fact-based, and free from hallucinations.
- PromptOps Infrastructure: Established a continuous evaluation and versioning pipeline (PromptOps) to iteratively improve the agent’s reasoning accuracy in complex diagnostic scenarios.
- Led .NET 8 upgrade project for 12 Commerce Integration Platform applications, completing ahead of .NET 6 end-of-life deadline
- Skilled in Agile development practices; actively contribute to Scrum meetings and collaborate effectively across all stages of the SDLC
- Implemented CI/CD pipelines and DevOps practices, improving deployment efficiency
- Conducted end-to-end API testing and managed infrastructure through ServiceNow ticketing system
- Gained exposure to data science workflows and cybersecurity practices through cross-team collaboration
- Teaching assistant for core CS courses: Artificial Intelligence, Data Structures, Programming Languages, Computer Organization
- Instructed 100+ students in machine learning algorithms, neural networks, and programming languages (Java, C, Haskell)
- Conducted lab sessions and office hours, improving student comprehension and course performance
Projects
Below are some of my key projects demonstrating my skills in machine learning, AI, and software development. Each project highlights my ability to tackle complex problems, implement effective solutions, and deliver impactful results.
Medical AI Data Pipeline & Expert System
- Engineered a comprehensive Medical AI system enabling unified processing of heterogeneous patient data (Excel & Medical Images) via a robust ETL pipeline. Delivered the full-stack MVP in 3 hours, demonstrating rapid prototyping capabilities.
- Rapid Prototyping: Delivered the full-stack MVP (ETL, Backend, UI) in 3 hours.
- RAG Engine: Integrated ChromaDB and LangChain for semantic retrieval and real-time analytics.
- Unified Processing: Enabled processing of heterogeneous patient data including Excel and Medical Images.
Medical-Expert Card System
- Built a generalizable physician experience card system to resolve "black box" and "hallucination" issues in medical LLMs. Implemented a "Visit-Trajectory-Knowledge" three-layer architecture for structured clinical data extraction.
- Three-Layer Architecture: Extracts VisitCard, TrajectoryCard, and KnowledgeCard from unstructured EHRs.
- Controlled RAG: Features Deterministic Indexing and Hybrid Retrieval for strict evidence citation.
- Algorithm-Guided: Utilizes prompt engineering to generate temporal disease progression trajectories.
Aortic Valve Insufficiency Diagnosis System
- Developed an automated diagnosis system combining 3D medical image segmentation and risk classification for improved early detection.
- 3D Segmentation: Utilized nnU-Net for precise segmentation of medical images.
- Risk Classification: Implemented XGBoost model for accurate risk assessment.
- Feature Analysis: Enhanced accuracy through radiomic feature analysis and morphological assessment.
Automatic Speech Recognition System
- Built an Automatic Speech Recognition (ASR) system using deep neural networks to improve transcription accuracy.
- Deep Learning: Implemented DNNs with CTC loss for acoustic modeling.
- Optimization: Applied Viterbi decoding and beam search for decoding optimization.
Cross-Modality MRI Synthesis
- Implemented CycleGAN architecture for medical image modality translation to enhance imaging workflows.
- Modality Translation: Achieved high-quality T1/T2 MRI translation.
- Workflow Enhancement: Enabled better analysis through synthetic data generation.
Netflix Data Wrangling, Exploratory Analysis, and Metadata Enrichment
- Performed comprehensive data wrangling, visualization, and enrichment on Netflix datasets to extract key business insights.
- Data Cleaning: Handled missing values, duplicates, and standardization using Pandas/NumPy.
- Visual Insights: Visualized trends and correlations with Matplotlib/Seaborn/Plotly.
- Data Enrichment: Integrated external ratings via OMDb API and performed feature engineering.