CSE @ IIT Tirupati | I build the infra layer things run on — ledgers, caches, simulators, agents
I like software that has to be correct, not just functional — things where a race condition, a lost message, or a bad cache invalidation actually costs someone money or data. That pulls me toward:
- Distributed systems & backend infra — consistency, idempotency, rate limiting, consensus, queues
- Systems programming — caches, memory hierarchies, pipelines, the stuff underneath the stuff
- Agentic AI — LLM-driven pipelines and tools that reason over code/data and take action, not just chat
Currently deep in distributed & scalable systems design, and picking up MLOps / continual learning on the side.
|
distriKV
A distributed key-value store built around Raft-based consensus for replication, an LSM-tree for persistence, and write-ahead-log recovery for crash safety.
|
|
aetherledger
High-throughput, multi-currency double-entry accounting engine. Immutable ledger entries, snapshot-based balance projections, Bellman-Ford-style FX routing, idempotency middleware, Redis-backed distributed sliding-window rate limiting, and a transactional outbox worker for reliable webhook delivery.
|
|
RISC-V-Simulator
Cycle-accurate RISC-V simulator modeling a multi-compute-unit architecture: shared fetch unit, two-level cache hierarchy (configurable size/associativity/latency), hardware-modeled |
ForgeCC-multibackend-compiler
A compiler for a custom statically-typed language, built from scratch in C with Flex/Bison. Full frontend (lexer, parser, semantic analysis, IR generation, optimization) feeding a retargetable backend — the same optimized IR is lowered to either a Java transpiler or RISC-V assembly.
|
|
nachos-project
Extended the NachOS teaching OS with real kernel features across dedicated branches: demand paging, a heap allocator ( |
gpumode-software
Frontend for a GPU-kernel benchmarking platform (KERN) — students submit GPU kernels which get judged and ranked. Handles JWT auth, async job polling for submission status, and a live per-question leaderboard.
|
|
truth-ahoy
A multi-agent misinformation-detection system. Google ADK agents orchestrate web search (Tavily), deepfake and reverse-image detection via Cloud Vision, and Vertex AI RAG retrieval over a BigQuery-backed corpus to separate fact from fabricated media.
|
bugSensei
A RAG-powered troubleshooting agent that turns vague error descriptions into precise, sourced fixes — built to skip the "Google it and hope" loop devs fall into when debugging.
|
|
PACGBI-Tool-Dev
Automated COBOL generation and refactoring pipeline driven off a GitHub issue backlog, using Mistral AI + GitHub Actions. Builds a code knowledge graph to prioritize high-impact changes and drive the agentic refactor loop end-to-end.
|
|
Dream11-InterIIT-TechMeet-13.0
ETL pipeline for raw fantasy-sports data feeding an XGBoost-based fantasy points predictor with best-XI recommendations (MAE: 140.6). Includes a train/test data selection tool and model-metric visualizations.
|
Adobe-InterIIT-TechMeet-14.0
AI-powered, Photoshop-style editor built for mobile creative workflows of the future — explores font-aware text editing directly on images and emotion-driven editing, as part of Inter-IIT Tech Meet 14.0.
|
Always down to talk distributed systems, backend architecture, or agentic AI — reach out.



