Developed a secure healthcare management platform with AI-powered medical document understanding for automated PDF parsing and structured data extraction.

• Express + MongoDB REST API (36 endpoints, 7 models) with dual JWT / Google OAuth authentication
• Cache-aside pipeline (Redis hot cache, MongoDB source of truth, 30-day TTL) with AI report analysis executed exactly once per record
• Automatic LLaMA → Gemini model fallback with fail-open behavior
• Cloudinary media storage and Multer upload handling
• 21 automated tests (Vitest, Supertest, in-memory MongoDB)

Next.js frontend with Express REST APIs and MongoDB document storage. JWT + OAuth 2.0 authentication flow, Redis-based cache-aside layer, and a LLaMA / Gemini medical entity extraction pipeline with reliable fallback.

36 REST endpoints · Redis cache-aside with 30-day TTL · 21 automated tests · Dual JWT/OAuth auth · Redundant de-risked document processing

Built an end-to-end clinical AI pipeline for coronary artery stenosis detection and segmentation using YOLOv8 models trained on 1,500 labeled coronary angiography images from the ARCADE dataset.

• Benchmarked YOLOv8n, YOLOv8s, and YOLOv8m variants
• Automated stenosis detection and instance segmentation
• Grad-CAM explainability for model interpretability
• FastAPI inference server exposing structured prediction APIs
• Automated COCO-to-YOLO annotation conversion pipeline

Dual YOLOv8 detection and segmentation models wrapped inside a FastAPI inference layer. OpenCV preprocessing pipeline with Grad-CAM heatmap generation for explainability and structured JSON response generation.

94.4% mAP@0.5 · 0.91 F1-score · Sub-200ms inference latency · Automated COCO-to-YOLO conversion saving 7+ hours of dataset preparation

Built a graph-native AI chat platform replacing traditional linear conversation history with interactive conversational graphs for non-linear thread exploration.

• Graph-based conversation visualization using React Flow
• Real-time thread persistence with MongoDB
• JWT authentication with Google OAuth 2.0 login
• Account deletion with cascading cleanup
• Context-aware AI interactions via OpenAI API

React frontend with React Flow graph rendering, Express API layer, MongoDB persistence, OpenAI API integration, JWT session management, and Dockerized deployment workflow.

Supports 10+ concurrent users · Non-linear conversation exploration · Automated NoSQL query sanitization · Improved context retention through graph navigation

Designed a fully serverless NLP analytics pipeline for large-scale review understanding and sentiment analysis using event-driven architecture.

• 4-class sentiment classification
• Key phrase extraction via AWS Comprehend
• Real-time review processing
• Event-driven architecture with CloudWatch monitoring

API Gateway → AWS Lambda → AWS Comprehend → DynamoDB → CloudWatch monitoring. Fully managed serverless architecture with independent service scaling and IAM-based access control.

Zero idle infrastructure cost · Fully serverless deployment · Independent component scaling · Automated review classification and insights extraction

Built a lightweight infrastructure monitoring system for automated Linux health tracking and maintenance using Bash scripting and Cron scheduling.

• CPU, memory, disk, and network monitoring
• Automated scheduled execution via Cron
• Threshold-based alerts for resource utilization
• Log compression and automated cleanup

Bash scripts orchestrated through Cron jobs. AWK and Grep used for metric extraction and parsing. Structured timestamped logging with automated maintenance workflows.

Automated collection of 6+ infrastructure metrics · Eliminated repetitive manual health checks · Automated log maintenance · Configurable resource threshold alerting

Developed a lightweight Linux security monitoring agent for centralized authentication and audit event collection with SIEM-ready output.

• SSH login monitoring and failed authentication detection
• Privilege escalation and session activity tracking
• SIEM-ready JSON event generation
• Sub-2 second event forwarding latency

Python-based event processing pipeline that parses Linux authentication and audit logs using regex and structures security events into JSON for downstream forwarding and analysis.

Processed 100+ daily authentication events · Sub-2 second event forwarding latency · Eliminated manual log inspection workflows · Centralized security event visibility