I’m a Computer Science undergraduate at Delhi Technological University (DTU), focused on building intelligent systems that bridge algorithmic thinking and real-world applications. My work spans machine learning research, computer vision, NLP, and GenAI systems — including RAG pipelines, agent workflows, and low-latency ML systems — with equal emphasis on correctness, scalability, and performance.
A lot of my repos come from a simple motivation: understand things deeply, then build them properly. That’s why you’ll find Transformers implemented from scratch, research papers re-created end-to-end, and systems designed with attention to stability, metrics, and failure modes and adversarial robustness.
Some things I’ve built here:
- BidNet — a low-latency real-time bidding engine optimized for scale and decision speed.
- SAR → EO Image Translation using CycleGAN — translating 2-band SAR imagery into 13-band multispectral EO images while preserving semantic and spectral consistency.
- LLM Safety Framework — red teaming pipelines, DistilBERT-based safety classifiers, and multi-strategy mitigation to reduce unsafe generations.
- Market Regime Detection System — unsupervised learning on high-frequency financial data using engineered microstructure features.
- InquireAI — a real-time RAG-based AI search assistant with streaming responses, reliable source ranking, and production-style backend design.
Alongside ML, I spend a lot of time on competitive programming and cybersecurity — not as separate hobbies, but as the same instinct: find the edge case, break the assumption, then build something that holds. This repo also hosts solutions to the CSES Problem Set, Advent of Code, and curated CP sheets, alongside CTF challenges I've designed — because strong fundamentals and offensive thinking make better ML engineers.
I enjoy working on problems where models fail quietly, data is messy, and performance constraints are real. Debugging unstable GAN training, reducing hallucinations in RAG pipelines, probing LLMs for adversarial inputs, or squeezing milliseconds out of inference paths - that's my comfort zone.
Outside code, I read late at night, overthink everything, and somehow lose hours scrolling reels that were supposed to last five minutes — occasionally pausing to think "I could've directed this better." I hit the gym to cope and storyboard scenes in my head that will probably never get made. Debug logs and emotions have similar energy.
This GitHub is my workspace — experiments, research implementations, system builds, and lessons learned the hard way. If something here sparks an idea, a critique, or a collaboration, I’m always open.
P.S. If any project made you think “Wait… this could actually ship 👀”, here’s my resume. Let’s build something meaningful.
