Projects

A real-time distributed tracing system built from scratch — visualizes live request flows across microservices with a D3 force graph and waterfall span view.

• Simulates 4 microservices (gateway, auth, orders, inventory) with configurable failure rates.
• Implemented distributed tracing using shared trace IDs propagated across service calls.
• Built a custom Prefix Trie for O(k) trace storage and lookup efficiency.
• Engineered a fixed-size worker thread pool for off-main-thread latency aggregation.
• Utilized WebSockets for real-time streaming of traces and metrics to the frontend.
• Visualized service dependencies using a D3.js force-directed graph.

A high-performance Least Recently Used (LRU) Cache implementation in C++ achieving O(1) time complexity for get and put operations.

• Architected using a hash map for O(1) lookup and a doubly linked list for O(1) eviction tracking.
• Implemented thread-safe operations using std::mutex to ensure data integrity.
• Includes comprehensive cache metrics (hits, misses, and hit rate) for performance monitoring.
• Built with a clean, modular design using CMake for build management.
• Included a custom benchmarking utility to validate performance requirements.
• Demonstrates fundamental systems programming concepts: memory management and data structure optimization.

An intelligent, multi-agent financial personalization engine built with LangGraph and Streamlit, designed to deliver high-conversion, context-aware nudges.

• Implemented a 3-node LangGraph pipeline for automated behavior analysis, RAG-based context retrieval, and LLM-driven message synthesis.
• Integrated ChromaDB and SentenceTransformers to maintain compliance with real-time SBI product guidelines.
• Engineered logic to handle complex financial profiles, ensuring correct interest rate calculations based on user demographics (Student vs. Senior Citizen).
• Created a high-fidelity mobile notification simulator within a custom Streamlit dashboard.

An AI-powered web research and report generation application built using Streamlit, LangChain, and Python, designed to transform natural language prompts into structured, insight-rich research reports.

• Designed a prompt-based interface allowing users to input any topic and receive a fully structured research report generated from real-time web data.
• Implemented an agentic RAG pipeline using LangChain, combining web search and scraping tools to retrieve and process relevant information from the internet.
• Integrated Tavily API for intelligent web search and BeautifulSoup-based scraping tools to extract and clean contextual information from multiple sources.
• Utilized LCEL (LangChain Expression Language) to build a modular and composable LLM pipeline for chaining prompt templates, models, and output parsers efficiently.
• Built a dual-agent system consisting of a search agent and a reader agent, enabling multi-step reasoning and improved information gathering.
• Leveraged Groq LLM (LLaMA 3.3 70B) to synthesize collected information into structured reports containing introduction, key findings, and conclusion sections.
• Implemented a writer–critic loop architecture, where a secondary LLM evaluates the generated report for clarity, depth, and factual quality.
• Enforced source-grounded generation to reduce hallucinations by restricting outputs to retrieved URLs only.
• Developed a modular and extensible architecture supporting scalable research workflows and future tool integration.

A full-stack movie recommendation and discovery web application built using Streamlit, FastAPI, and Python, designed to deliver personalized cinema insights powered by machine learning and NLP techniques.

• Designed an interactive and cinematic UI using Streamlit to enable seamless exploration of movies through search, genres, and trending categories.
• Implemented a content-based recommendation system using TF-IDF vectorization and cosine similarity, enabling users to discover similar movies based on titles, genres, and keywords.
• Applied NLP text preprocessing techniques (such as cleaning, normalization, and feature extraction) to improve text representation and recommendation quality.
• Integrated TMDB API to fetch real-time movie metadata including posters, ratings, descriptions, and popularity trends.
• Built a scalable backend using FastAPI, deployed on Render, to efficiently serve recommendation and movie data through REST APIs.
• Utilized pickle-based model persistence to store precomputed similarity matrices and vectorized features for fast inference.
• Optimized performance using NumPy and Pandas for efficient computation, ensuring low-latency recommendations and smooth user experience.

A responsive and modular personal portfolio website built using HTML, CSS, and JavaScript to showcase projects, skills, achievements, and professional experience.

• Designed with a clean and intuitive user interface, ensuring consistent styling, smooth navigation and clean layout transitions across sections.
• Implemented smooth scrolling, interactive elements, and animations to enhance user engagement and browsing experience.
• Developed a fully responsive layout optimized for desktops, tablets, and mobile devices.
• Developed a fully responsive and mobile-friendly website optimized for all devices, with a modular code structure leveraging reusable components and organized code to ensure maintainability and scalability.