I gravitate toward the hard parts of problems: scale, structure, and signal in the noise. I build systems from scratch, measure them rigorously, and tune until the numbers move.
I'm Ishita Tyagi, a Pre-Final Year CSE student at IIT Jodhpur. Most of what I build lives close to the metal: a chaos-tested distributed-systems simulator, a version-controlled SQL engine with Kafka-backed rollbacks, a peer-to-peer network with exactly-once delivery, and real-time computer vision squeezed onto a Raspberry Pi.
My interests sit at the intersection of systems, applied machine learning, and data-driven analysis. I value clean abstractions, measurable performance, and learning by building. Now in the penultimate year of my B.Tech, with the internship window open, I'm looking to take that same rigor — whether building systems or analyzing them — to problems that run at real scale.
The languages, systems, and infrastructure I reach for most.
Complex systems designed for production and experimental use, built from scratch, measured, and tuned.
A from-scratch C++ engine that models distributed architectures (microservices, peer-to-peer, and IoT) across 10+ dynamically allocated nodes. A configurable chaos-testing framework injects 5+ real failure modes and forecasts P50/P99 latency and SLO violations (<5%) over 100+ scenarios, using custom event queues and topology-aware routing with no external simulation libraries.
A version-controlled SQL workbench whose rollback engine versions 100% of write operations through inverse-query generation across concurrent multi-user sessions. A Kafka pipeline cut rollback latency by ~60% under 50+ concurrent requests, Redis serves sub-millisecond reads, and AWS S3 storage reduced costs by ~75%, a production-grade approach to safe, auditable database tooling.
A full-stack, multi-device platform that auto-categorizes activity using 384-dimensional transformer embeddings with PostgreSQL pgvector HNSW search, with no manual tagging. Real-time sync across an Android app and a Chrome MV3 extension feeds a Node/Express server with WebSocket streaming, JWT auth, and a Glicko-style Bayesian rating engine: AI-integrated software built to run continuously at scale.
A decentralized peer-to-peer network built from raw TCP sockets and multithreading, with SHA-hash deduplication guaranteeing exactly-once delivery. A Barabási–Albert scale-free topology with PING/PONG liveness evicts dead peers in under 14s and stays stable through 30% node churn via thread-safe, synchronized connection management.
Real-time gesture recognition and object detection running on a Raspberry Pi, made deployable through model pruning and 8-bit integer quantization that cut inference latency by ~30%. Pipeline parallelism drives autonomous navigation, with camera calibration and confidence-threshold tuning for reliability under noisy, real-world lighting.
A real-time C++/SFML visualizer that races five pathfinding algorithms (BFS, DFS, Dijkstra, A*, and Greedy Best-First) across randomly generated mazes, rendered step-by-step. A clean OOP design (an abstract Solver base class with per-algorithm subclasses) feeds a benchmark screen comparing runtime, nodes explored, and path length.
A small programming language built from scratch in C with a Lex-based lexer and a Yacc grammar. Source is tokenized into a typed stream (identifiers, operators, numerics), then parsed and evaluated, implementing the classic compiler front-end: lexical analysis, grammar rules, and semantic actions.
Engineering write-ups and reflections, publishing soon.
Securing 3rd place in the individual Student Academic Conference event: presenting work, fielding tough questions, and competing against students from across the IITs. Full write-up coming soon.
Computer Science undergraduate at IIT Jodhpur, open to SWE, analyst, and research internships for Summer 2027.