Production-grade GitHub App that listens for pull request webhooks, performs context-aware AI code review, and posts structured findings directly as inline PR comments. Implements JWT-based GitHub App authentication with installation token caching, HMAC-SHA256 webhook signature verification, and diff-chunking to stay within LLM context limits while preserving surrounding code context. Uses OpenAI's structured JSON output mode to guarantee parseable findings (severity, category, file, line, suggestion), and a React dashboard with MongoDB aggregation pipelines to surface code-quality trends across repos over time.

Node.js Express MongoDB React OpenAI API

Real-time load balancer simulator implementing 5 pluggable scheduling strategies (Round Robin, Weighted RR, Least Connections, Least Response Time, Predictive Headroom Forecasting) via the Strategy design pattern. Built a custom indexed min-heap for O(log N) dynamic priority routing, and WebSocket telemetry with backpressure-aware frame dropping to prevent buffer bloat under high-frequency updates. Predictive scheduling reduced load-distribution standard deviation from 15.72 to 3.15 versus Weighted Round Robin.

Node.js TypeScript React WebSockets Recharts

Built production-style automation workflows in n8n, including an email-summarization pipeline and an LLM-powered auto-reply system for sponsorship/outreach emails, cutting manual response time by an estimated 70%. Applied function-calling patterns to connect LLMs with external tools (Google Sheets, webhooks, APIs), and am extending this into RAG-based agents — vector embeddings, retrieval-augmented generation, and chunk/overlap optimization — for knowledge-retrieval chatbots with source-grounded responses. Working across both closed APIs (OpenAI) and open-source models (DeepSeek, Llama, Mistral).

n8n LLM APIs RAG Vector Databases Prompt Engineering

A structured, pattern-based approach to DSA mastery — rather than grinding problems topic-by-topic, organizing practice around 28 core algorithmic patterns that cover the majority of interview problems, following a self-imposed engineering discipline: a 30-minute struggle rule before consulting solutions, brute-force-to-optimal iteration, and every solution tagged with its underlying pattern and a written "Aha!" insight. Structured roadmap spans two-pointer/sliding-window fundamentals through trees, graphs, and dynamic programming, in C++20.

C++ Algorithms Data Structures Problem Solving