Projects

Receipt Processor

Built a RESTful API using Python and FastAPI to process JSON receipt data and compute reward points using rule-based logic, with modular design separating routing, validation, and business logic
Enforced strict input validation using Pydantic with regex constraints and custom error handling for consistent and informative validation responses and reduced invalid request processing by 100%
Containerized the application using Docker and designed the service to be stateless, scalable, and easily testable with integration tests written in Pytest, increasing test coverage by 60%

Conceptualized an open-source web app using React, Firebase Auth, Firestore DB to address the need for a simple, secure and scalable task manager
Implemented secure authentication and used sub-collections, ensuring users’ tasks persist and remain isolated
Enhanced the UI with Bootstrap, enabling intuitive features like creating, reordering, and deleting tasks efficiently
Deployed via Vercel, achieving sub-second load times, broad accessibility and laying the groundwork for future expansions

Designed a schema & developed a suspicious transaction detection and flagging dashboard using SQL Server and Power BI to identify anomalies and fraudulent transactions
Optimized data processing and reporting using indexing and views to pre-aggregate frequently used metrics, improving query performance by 20%
Implemented percentile-based categorical transaction bucketing, duplicate checking using SQL to enhance anomaly detection and risk analysis

Created an N x N grid environment with stochastically moving ghosts, randomly spawned walls and our agent.
Implemented various search algorithms (DFS, BFS, BDBFS, A-Star etc.) to plan a path for the agent from start to finish while avoiding ghosts and walls and analyzed their performance.
Used various heuristics, some standard and some novel.

Simulated a circular graph environment with 3 entities-Agent, Prey, and Predator. The Prey moved probabilistically, while the Agent was optimized to catch it, and the Prey moved greedily toward the Agent.
Implemented Markov Chains, designed a custom Neural Network and compared them to optimize the Agent’s decision-making, reducing the time to catch the Prey by 5%

Led a team of 3 developers to design and implement a web-based auctioning platform with features like real-time bidding, auto-bidding and bid notifications to improve user engagement.
Coordinated full-stack development using MySQL and Java for the back-end, and JSP for a dynamic front-end, achieving a 15% improvement in response times compared to the initial prototype through query optimization and asynchronous updates
Delivered the project within a 3-month academic timeline, demonstrating the ability to handle up to 100 concurrent bids for test users

Implemented a custom TCP kernel module for Linux from scratch, achieving 100% throughput under test conditions
Designed & integrated a novel congestion control algorithm variant, improving throughput by 35% over TCP-Reno
Managed low-level packet transmission, ACK handling, and window scaling logic in the Linux kernel networking stack
Conducted extensive benchmarking & validated <1% packet loss and RTT stability
Debugged complex kernel-space issues, reducing crash frequency