Projects & Innovations

A production-grade web application designed to "decode" any GitHub repository. It transforms raw code into a structured, visual "Code Wiki" with instant technical documentation and high-fidelity architecture diagrams. 🏗️ Architecture & Structure: • Core UI Layer: Manages application lifecycle, Wiki-style deep links, and complex analysis state. • Service Layer: - githubService.ts: Data acquisition engine with recursive tree fetching and multi-tier fallback (API → JsDelivr → Raw). - aiService.ts: Deep-context analysis using Gemini 3.1 Pro focusing on "Why" and "How". - graphEngine.ts: Custom static analysis parsing file relationships into Mermaid.js diagrams. • Smart Context Injection: Prioritizes "DNA files" (READMEs, manifests, entry points) for high signal-to-noise ratio. • Resilient Fetching: Fallback strategy to ensure 100% uptime and bypass GitHub API rate limits.

A B2B SaaS platform transforming Healthcare RCM through autonomous data auditing and predictive analytics. 🚀 Project Overview: • Autonomous Agent Architecture: Specialized AI "Pods" (CEO, Engineering, Data/ML, DevOps) collaborate to build and scale. • Automated Data Auditing: Autonomous scanner identifying billing errors and missing data with custom heuristic logic. • Predictive Payment Forecasting: Scikit-learn regression models (Random Forest) to predict cash flow timelines. • CRM Integration: Direct API integration with Monday.com for seamless operational synchronization. • Infrastructure: FastAPI backend, React/Vite frontend, Docker orchestration, and GCP deployment.

An advanced career development platform automating the end-to-end job-seeking process using a decentralized multi-agent approach. 🏗️ Technical Architecture: • Agent Alpha (Resume Architect): Multimodal analysis using "Thinking" budget for deep candidate impact reasoning. • Agent Bravo (Market Scout): Real-time salary and trend fetching using Google Search Grounding. • Interview Simulator: Low-latency, full-duplex voice interactions using Gemini 3.1 Flash Live API and Web Audio API. • Coding Lab: Real-time code evaluation for time/space complexity and style. • UI/UX: High-performance "Glassmorphism" interface with custom SVG-based data visualizations.

Enterprise internal platform for a healthcare RCM organization to centralize onboarding, training, task management, and admin reporting. Technical Highlights: • Hybrid Database: PostgreSQL (SQLAlchemy) for relational data and MongoDB (GridFS) for scalable document storage (PDFs, PPTs). • Workflow Logic: Training-and-progression system with company email validation, 2FA, approval gates, and timed exams. • Task Management: Admin-assigned client tasks with status milestones, automated notifications, and scheduled cleanup via APScheduler. • Enterprise Security: CSRF protection, Redis-based rate limiting, account lock logic, and secure session handling. • DevOps: GitHub Actions CI/CD with Azure Web App deployment and Flask-Migrate for schema management.

Comprehensive research on applying ML to forecast payment behaviors and optimize revenue flow in healthcare. 🎯 Research Objectives: • Develop models for payment likelihood, provider variations, and location-based trends. • Analyze key drivers influencing payment behavior (Days since billing, Payer mix, Previous behavior). 📊 Methodology: • Dataset: 3-year HIPAA-compliant dataset (~28,000+ records). • Feature Engineering: Payment history lags, rolling averages, and Branch & Bound inspired feature selection. • Models: Compared Linear Regression, Decision Trees, Random Forest, and Neural Networks. 🔍 Key Findings: • Future Payment: Random Forest & Neural Networks achieved R² > 0.90. • Revenue Forecasting: Simpler linear models outperformed complex ones due to bias-variance tradeoffs on limited time-series data. • Location Trends: Used K-Means clustering to segment facilities and improve prediction accuracy.

End-to-end AI platform analyzing ~7M+ Yelp reviews to generate actionable sentiment insights and recommendations. 🏗️ System Architecture: • ETL Pipeline: Multi-threaded batch ingestion from raw JSON to PostgreSQL with dynamic schema filtering and foreign key validation. • Feature Engineering: - Text: VADER sentiment scores and TF-IDF (Top 50 terms). - Temporal/Geo: Weekend/holiday indicators and Geo clustering (K-Means). - Behavioral: User retention modeling and business review density distribution. • Machine Learning: Ensemble VotingClassifier (Logistic Regression, RF, XGBoost) with Soft Voting and Stratified K-Fold validation. • LLM Layer: Design-ready architecture for review summarization and pros/cons generation.

Critical incident response system designed to mitigate the impact of floods in Sri Lanka through real-time mapping and mass communication. Key Capabilities: • Mass Alerting: Integrated SMS gateways for automated emergency alerts to at-risk populations based on geo-fencing. • Incident Mapping: Real-time location tracking for flood reports and emergency resource allocation using mapping APIs. • Crisis Coordination: Centralized dashboard for disaster management units to coordinate field responses and monitor weather behavioral trends. • Impact: Provided a scalable solution for national-level emergency response coordination.

Interactive web application for comprehensive statistical analysis and machine learning. Features: • Data Processing: Robust cleaning tools for missing value handling (Mean/Median/Custom), outlier detection (IQR/Z-Score), and scaling. • Inferential Statistics: Interactive T-Tests, ANOVA, Chi-Square, and non-parametric alternatives with full visualization. • ML Lifecycle: Full support for Regression, Clustering (K-Means, DBSCAN), and PCA dimensionality reduction. • Dashboard: Dynamic visualizations using Plotly and Seaborn with interactive filtering and data export (CSV/Excel).