Software engineer · Systems × GenAI · M.Tech, IIIT Delhi
Explore my portfolio & project case studies ↗ / Projects / Merged contributions / LinkedIn
I build across systems engineering and generative AI—from C++ linker fixes and Linux concurrency to retrieval-backed assistants and LLM-powered developer tools.
I care about what happens around the core algorithm: reproducible failures, useful evidence, reliable APIs, and a product people can inspect and use.
At IBM India Software Labs, I worked on the IBM Z C/C++ compiler, resolved 8+ production defects, and implemented an MCP service for compiler-diagnostic retrieval. Software Engineer: July 2025–September 2026; intern: January–July 2025.
My portfolio brings together five featured projects, linker contributions, implementation details, and code links. Search by technology or explore systems, AI/ML, applications, and research.
Five featured projects, in the same order as my portfolio.
 Two merged fixes in an ELF linker built on LLVM:
Focused regression testing across six architecture configurations. Each PR records the verification scope.
|
 Contributed hybrid BM25 + MiniLM retrieval over a 1,871-page historical IPC PDF, with reciprocal rank fusion, page-linked evidence, and cited research briefs. Local scope/intent classification and a RAG assistant with document-excerpt fallback.
|
 A PHP/Slim property marketplace connecting searchable listings, account workflows, MySQL persistence, Razorpay integration, and generated PDF contracts. Full-stack prototype
|
 A notebook experiment fine-tuning GPT-2 to turn Amazon Fine Food Reviews into short summaries, covering data preparation, training, checkpointing, and generation. ML coursework · evaluation corrections documented
|
|
Interactive C++ lessons and a C++17 filesystem lab exploring containers, recursive operations, and concurrency.
Two complementary labs: browser teaching simulations and an educational filesystem interpreter. Includes timing scripts; no measured speedup claimed. Live demo ↗ · Concepts source ↗ · Systems source ↗ · Case study → |
|
Earlier projects & explorations
More prototypes and learning projects, preserved in the portfolio archive. Illustrations below are concept artwork.
 Compiler investigation workspace with model-assisted routing, isolated LLVM recipes, SQLite history, resumable events, and hashed evidence exports. Status: synthetic replay demo. Live model/worker integration validation pending.
|
 RSS ingestion, deduplication, and heuristic ranking feed a script-drafting workspace with optional Gemini/OpenAI generation, saved articles, and text/PDF exports. Status: local prototype with a template fallback. Live-provider validation and deployment hardening are pending.
|
 A native Windows companion that combines an animated desktop pet with focus sessions, local activity history, persistent streaks, and a productivity dashboard. Local Windows application
|
 An event-driven multiplayer quiz with host-controlled rooms, QR-code joins, timed questions, server-side scoring, and live leaderboards. Web application · room state held in one server process
|
 A related pair of research tools: an extremophile literature-mining pipeline and an interactive chemical–gene graph that makes prepared relationships and publication references easier to explore. Research capstone · two connected repositories
|
 An Android app that runs a bundled quantized MobileNet model locally to classify a selected gallery image and display its predicted label. Educational mobile prototype
|
 A Flask image-captioning prototype combining InceptionV3 visual features with an LSTM language decoder to turn an uploaded image into a caption. Earlier ML prototype
|
 An educational PHP toolkit exploring hash demonstrations and basic website configuration inspection through browser-based forms and result panels. Educational prototype for authorized lab use
|
A recent example: ELD #2134 — preserve AArch64 feature flags. A one-line C++ fix, a regression that failed before the change, and 18 passing focused test runs across two AArch64 configurations. PR open for review.
AI tooling: Hugging Face smolagents #2949 — Draft AST-validation fix for unpacking aliases in context managers, with tool export/reload regression coverage. The PR records 70 focused pytest passes; scope acceptance and contributor review are pending.
More open-source contributions and verification scope
- ARM/AArch64 LTO fix · #2129 · Open — Preserve native object generation with
--save-temps. Full LTO, ThinLTO, and explicit assembly-output coverage; 16 focused tests passed, followed by the updated regression on all six target configurations. - Build guidance · #2127 · Open — Document the LLVM
mainbaseline and source/build path guidance following a maintainer discussion. - React ChatBotify · #366 · Open — Submitted chat-history HTML sanitization, URL checks, and an optional sanitizer callback with tests.
Statuses checked 13 October 2026. ELD #2011 and #2065 are merged; #2134, #2129, and #2127 remain open. smolagents #2949 is a draft; React ChatBotify #366 remains open. Results above describe focused tests recorded in the PRs, not full-suite runs or live integration validation of every project.
Algorithms and coding practice
DSA · CSES · LeetCode daily work · A-to-Z series · Earlier LeetCode practice · Striver A-to-Z
| Systems engineering | GenAI & applied ML | Backend & delivery |
|---|---|---|
| C / C++, Linux, GDB | RAG, BM25, MiniLM, RRF | Python, FastAPI, SQL |
| LLVM, CMake, Ninja, lit | LLM APIs, PyTorch, Hugging Face | TypeScript, Next.js, Docker |
| Filesystems, concurrency | TensorFlow Lite, scikit-learn | PySide6, PHP, MySQL |
Let's build at the intersection of systems and AI. Visit my portfolio ↗ · Connect on LinkedIn →