★ Featured
openSc0ut - Find, Contribute, Repeat.
A real-time AI chat platform powered by fine-tuned LLMs with retrieval-augmented generation.
Next.jsTypescriptTailwindPyTorchLangChainRedis+1
Skills & Tools
My Tech Stack
Technologies I work with daily, spanning frontend, backend, AI/ML, and DevOps.
Portfolio
Selected Projects
Handpicked work that demonstrates technical depth, problem-solving, and real-world impact.
★ Featured
Orion — 3D Data Visualization Engine
A WebGL-powered analytics dashboard for exploring large-scale graph datasets in 3D space.
ReactThree.jsTypeScriptD3.jsGraphQLNode.js
ContextOS — Semantic Search Engine
A developer-first semantic search engine with vector embeddings and hybrid retrieval.
PythonFastAPIPyTorchFAISSDockerReact
Career
Professional Experience
Where I've worked, what I built, and the impact I delivered.
Research Engineering Intern
DeepMind —
3 moLondon, UK (Remote)Worked on scalable training infrastructure for large language models within the Gemini research team.
- Optimized distributed training pipeline reducing GPU idle time by 18% across 512-GPU clusters
- Implemented a novel attention masking strategy that improved long-context coherence on internal benchmarks
- Contributed to internal tooling for experiment tracking and hyperparameter sweep automation
Software Engineering Intern
Razorpay —
2 moBangalore, IndiaBuilt and shipped features for Razorpay's payment analytics platform serving 8M+ merchants.
- Redesigned the refund reconciliation service, cutting processing time from 12 hours to 45 minutes
- Developed a fraud signal aggregation API handling 500k events/day with <10ms p99 latency
- Shipped a new merchant dashboard cohort analysis feature adopted by 30% of enterprise clients
Undergraduate Research Assistant
IIT Bombay — ML Lab —
1y 9moMumbai, IndiaConducted research on multimodal learning and efficient inference for vision-language models.
- Co-authored a paper on speculative decoding for VLMs accepted at EMNLP 2023
- Built an open-source benchmark suite for cross-modal alignment, used by 15+ research groups
- Mentored 4 junior researchers on deep learning fundamentals and research methodology
Publications
Research Papers
Academic work at the intersection of NLP, vision-language models, and efficient inference.
EMNLP 20232023
SpectralDraft: Speculative Decoding for Vision-Language Models via Token Spectrum Analysis
We introduce SpectralDraft, a speculative decoding framework tailored for vision-language models. By analyzing the token probability spectrum across modalities, our method selects high-confidence draft tokens that achieve a 2.8× speedup over autoregressive baselines with negligible quality degradation on VQA and captioning benchmarks.
Preprint — arXiv 20242024
CrossAlign: A Benchmark for Evaluating Cross-Modal Semantic Alignment in Foundation Models
CrossAlign introduces a systematic benchmark for measuring semantic alignment across vision, language, and audio modalities in large foundation models. We evaluate 12 state-of-the-art models across 8 alignment tasks, revealing significant gaps in compositional reasoning that current training objectives fail to address.
Get in Touch
Let's Connect
Whether you have a project in mind, a research collaboration, or just want to say hi — I'd love to hear from you.
Let's work together
I'm currently available for freelance projects, research collaborations, and full-time opportunities.