yeshwanth s.

CS Student @ PSGCT

I explore applied AI, databases, and backend systems.

I like understanding how things work, then turning that understanding into tools people can use.

I work with Python and C++ — sharing what I’m building on and . I love exploring cool ideas and experimenting with new tech. To collaborate, say hi.

Tamil Nadu, India

selected work

View all
Hotpath application interface
HotpathLearn AI engineering from the inside out. Build model internals, run Python labs, and make working code faster.Next.js · Python · AI engineering
zug
××××××××××××××××××××××××××××××××××××××××××××××××××××××××
120 cells. 64 squares.A board that knows its boundaries.
Chess, from first principles.C++17 ↗
ZugA chess engine built from first principles. Every legal move, pruned branch, and undone position written by hand.C++17 · Alpha-beta · UCI
Vigilens application interface
VigilensReal-time video intelligence for always-on RTSP streams with fast screening, Gemini verification, durable SQLite memory, and searchable query history.Python · FastAPI · Gemini API

About

I like taking ideas apart, understanding the hard parts, and turning that understanding into tools people actually use.

Along the way.

May 2026 – Jun 2026IIT Madras, WSAIResearch Intern

Conducted research on accelerating Vehicle Routing Problems with Time Windows (VRPTW) using Graph Neural Networks within a Column Generation framework.

Inside the work
  • Studied and evaluated a Bipartite Graph Neural Network for route selection within a VRPTW Column Generation framework.
  • Performed extensive calibration experiments using threshold tuning and class-weight optimization, significantly improving Balanced Accuracy and True Negative Rate.
  • Integrated Weights & Biases Bayesian Sweeps to optimize hyperparameters and positive class weighting.
  • Designed and evaluated a CTGAN-based synthetic route generation pipeline.
  • Conducted comprehensive ablation studies to analyze the contributions of graph topology, message passing, and route-level features.
  • Implemented and evaluated SPPRC optimizations, including 2-cycle elimination.
  • Integrated the optimized GNN into the Column Generation pipeline and benchmarked computational efficiency.
2023 – PresentPSG College of TechnologyUndergraduate Student

Studying Computer Science and Artificial Intelligence and Machine Learning.

June 2024 – May 2026THE EYE (Cybersecurity & AI Society)Core Team Member

Built and shipped backend and interactive tools for events, workshops, and community projects.

Inside the work
  • Co-developed a backend CTF portal with authentication, a WebSocket scoreboard, challenge unlocking, and admin tooling that scaled to 300+ players.
  • Built TranscribeX, a workshop-deployed web app for video transcription, subtitle generation, and AI summarization workflows.
Higher SecondaryBharathiyar Matric Hr. Sec. SchoolHSE - 98.5%

Completed higher secondary education with 98.5%.

The toolkit.

Languages

3 tools
C++GoPython

Backend & DB

7 tools
FastAPIPostgreSQLMongoDBRedisSQLiteConvex DBSupabase

Infra & Tools

7 tools
DockerGoogle CloudGitLinuxVS CodeVercelPostman

AI & ML

7 tools
Hugging FacePyTorchPandasVertex AIOpikOllamaMLflow

Outside the usual.

Notes from the build

Zug: building a chess engine from first principles

Why a 120-cell board, reversible moves, and a simple evaluator were enough to begin.

vigiLens - Always-On Video Intelligence

A deep dive into building vigiLens: real-time screening, Gemini-powered verification, SQLite event memory, and natural language query retrieval for RTSP streams.

have an idea?
let’s talk.

a project, a question, or just a hello — my inbox is open.

say hello