I build what I wish existed.

Projects

01

Learn That Song live

Paste a YouTube link, get six instrument stems you can mute, slow down, transpose and loop. CloudFront serves the page, which sends the link to a Lambda API that queues a job in SQS. An on-demand EC2 worker splits the audio with Demucs into S3, and the browser caches the stems and plays them in sync with Tone.js.

Python · JavaScript · PyTorch · Tone.js · AWS (Lambda, SQS, EC2, S3, CloudFront)

Try it live →GitHub →

02

CNN-Based Local Feature Descriptor Learning from a Single Image

A CNN that turns patches around keypoints into feature descriptors, trained without a large dataset. Crops and augmentations of a single source image are paired so the network learns which patches show the same point. Keypoints from OpenCV are described, matched across image pairs and benchmarked against SIFT: within 1.5% of its accuracy over 70 pairs, with 4% more matches.

Python · PyTorch · OpenCV

GitHub →

Experience

1. Jul 2025 – Oct 2025 · Internship · Canberra

   Trellis Data

   AI Programmer Intern

   At ML Research Labs, Trellis Data's R&D division, I worked on an early-stage LangGraph agentic translation workflow for preserving Indigenous languages, inspired by the Machine Translation from One Book paper.

   • Improved translation accuracy by 20% with a new chunking strategy.
   • Improved translation quality by 8% by rewriting LLM queries to target recurring errors, such as missed pronouns.
   • Made load times over 80% faster by switching to persistent FAISS vector stores.
   • Built a FastAPI interface so external users could interact with the workflow.
   • Cut evaluation time by 36% by parallelising 20+ Gemini mixed-model setups with asyncio, keeping translation quality within 3%.
   • Python
   • LangGraph
   • FastAPI
   • FAISS
   • Gemini
   • asyncio
2. Aug 2021 – Jul 2024 · Full-time · Bangalore

   Hewlett Packard Enterprise

   Software Engineer

   In HPE's HPC division, I worked on the System Monitoring Application (SMA), a distributed monitoring and diagnostics application used by several of the world's top supercomputers.

   • Extended SMA's Kafka-based data collection services and built visualisation dashboards in Grafana.
   • Tracked down issues in production test environments ahead of each release.
   • Picked as the sole representative from a global team of senior engineers to fly to LUMI, Europe's largest supercomputer, and fix an issue blocking acceptance.
   • Over one week on-site, found the data filtering setup that kept critical monitoring data flowing without gaps under a tight ingestion limit, and saw SMA through to sign-off.
   • Redesigned 50+ Grafana dashboards with the on-site hardware engineers, and made them up to 80% faster to load by optimising the underlying SQL queries.
   • Python
   • Go
   • Linux
   • Kubernetes
   • Kafka
   • PostgreSQL
   • Elasticsearch
   • Logstash
   • Grafana
   • Helm
   • Podman
   • Docker
   • Ansible
3. Feb 2021 – Jul 2021 · Internship · Bangalore

   Hewlett Packard Enterprise

   Software Engineer Intern

   I started at HPE as an intern in its largest R&D division in India, on a data centre digital transformation initiative to replace fragmented Excel-based tracking.

   • Built a full-stack MERN inventory management application with a three-tier RBAC approval workflow.
   • Added in-app migration of legacy data from FY 2019-20 and FY 2020-21, with MongoDB schema and Formik validation flagging malformed values.
   • Enabled real-time monitoring of 5,000+ consignments a year.
   • Turned the internship into a full-time Software Engineer offer, with a recommendation from my mentor.
   • JavaScript
   • Express.js
   • React.js
   • Node.js
   • MongoDB
   • Formik
4. Jun 2019 – Jul 2019 · Internship · Mumbai

   Fourth Signal

   Web Development Intern

   My first internship was in Fourth Signal's web development team, enhancing their in-house, client-facing platform.

   • Refined the methodology for a quantitative mutual fund risk-score model with my mentor, ranking 50+ debt and hybrid funds.
   • Implemented the model and its front-end components as a fund-ranking and recommendation feature.
   • PHP
   • MySQL
   • Bootstrap

Off the clock