Content
AI Healthcare System
| **AI Healthcare System | Python, PySpark, Databricks, Delta Lake, Airflow, PostgreSQL, Docker, RAG** |
A production-grade healthcare data platform designed for scalable ETL processing, dimensional modeling, and low-latency clinical retrieval.
Tech Stack
- Big Data & Lakehouse: PySpark, Databricks, Delta Lake (Medallion Architecture), SCD Type 2
- Orchestration & Database: Apache Airflow, PostgreSQL
- Backend & Containerization: FastAPI, Docker
- GenAI / Search: RAG pipeline, Cloudflare AI embeddings, Local Vector Cache
Key Highlights & Results
- Lakehouse ETL & Modeling: Designed Batch and Streaming ETL workflows handling 1M+ records using Delta Lake Medallion Architecture and SCD Type 2 modeling, optimizing partitioned storage and MERGE operations to reduce runtime by 40%.
- High-Performance Serving: Delivered FastAPI endpoints and PostgreSQL indexing to lower analytical query latency by 35%, containerized workloads with Docker, and scheduled jobs using Airflow DAGs.
- Optimized RAG Architecture: Implemented a Retrieval Augmented Generation pipeline integrating Cloudflare AI embeddings with a local vector cache, compressing context payload size by 80% and achieving sub-100ms response times.
System Architecture
graph TD
User([User / Patient]) -->|Vitals & Queries| App[Streamlit App]
App -->|API Request| Backend[FastAPI Backend]
subgraph Engine [AI & Analytics Engine]
Backend -->|Extract Lab Reports| GeminiVision[Gemini Vision API]
Backend -->|Run Symptoms Classifier| XGBoost[XGBoost ML Models]
Backend -->|Semantic Search| FAISS[(FAISS Vector DB)]
FAISS -->|Retrieve Context| GeminiPro[Gemini Pro RAG]
end
XGBoost -->|Diagnostic Prediction| Backend
GeminiPro -->|Grounded Medical Assist| Backend
Backend -->|Response Payload| App
