Software and Machine Learning Engineer. I turn research into software people can rely on.
New Delhi, India · Computer science at IIIT Delhi · Graduating September 2026
Portfolio · LinkedIn · Email · SDE résumé · DS résumé
| Role | Where | When | What |
|---|---|---|---|
| Head of Product Development | Student Council, IIIT Delhi | 2025 to present | Built and shipped a student services portal used by more than 500 students; own the database, the indexing, and the API eight engineers build against. Set up CI with automated tests and review the team's code. Replaced manual data entry with validating forms, cutting admin work by about 60 percent and stopping a recurring class of bad records. |
| Software Engineering Intern | Shipzee | Mar to Aug 2026 | Shipped features on a live courier aggregation product, from the React screens through the API to the database. Led a move to a new stack planned so nobody saw an outage. Traced failures that crossed service boundaries through logs and request traces, and wrote contract tests around external integrations. |
| AI Engineer and Undergraduate Researcher | MIDAS Lab, IIIT Delhi | May 2025 to Jan 2026 | Built an evaluation harness that scores model output against several notions of alignment and flags on its own when quality drops. The checking step cut fabricated answers by about 20 percent and the lab still runs it. Reproducible pipelines over three datasets with retry and rate limit handling. Trained models with LoRA and QLoRA at four bit precision. First author on a paper currently under review. |
| ML Research Intern | Computational Social Science Lab, IIIT Delhi | Jan to May 2025 | Built a parallel Google Earth Engine pipeline producing district scale multi channel inputs and segmentation masks. Compared CNN backbones on phenology, crop type, and yield under one shared setup. Measured drift across more than ten districts with difference in differences and fixed effects models. |
| Data Science Intern | Complex Systems Lab, IIIT Delhi | Aug to Dec 2024 | Led a team of five on a 25 class classifier over messy real world data, reaching 84.5 and 74.5 percent on two splits, and traced the gap to a shift in the data. Built web apps that visualised complex datasets in real time. Ran reviews and mentored the junior members. |
| Open Source Contributor | Meshery, Layer5 | 2025 | Bug reports, feature requests, documentation fixes, and pull requests; took part in code reviews. |
| Project | Type | Summary | Stack |
|---|---|---|---|
| Executive Career Agent | Agent swarm · desktop | Seven cooperating agents discover, deduplicate, score, and tailor roles against a candidate's real résumé. Runs fully offline or against the live web; a validator checks every claim against the master résumé. Ships an MCP server for Claude Desktop. | React · Electron · Node.js · MCP · LangChain |
| TURBO | ML research | Independent reimplementation of Goel et al. (arXiv 2502.07391): fuses Vision Transformer image features with a modified BART text encoder for multimodal sarcasm explanation, trained on MORE+ and scored with BLEU, ROUGE, and METEOR. | PyTorch · ViT · BART · Hugging Face |
| MineMEETS | Multimodal retrieval | Turns images, text, and audio into embeddings behind one interface. Every step is dimension checked and safe to rerun; Pinecone namespaces make reindex and rollback safe; tests and type checks run on every PR. | CLIP · Whisper · Pinecone · Docker · Pytest |
| Cross Portfolio Default Risk | Data science research | Dual study of credit default prediction across credit cards (UCI, 30K) and installment loans (LendingClub). 31 engineered features, 13 calibrated models with SHAP and cost sensitive thresholds. XGBoost reached ROC AUC 0.7769 on cards, about $18.7M estimated annual savings. | XGBoost · LightGBM · SHAP · Optuna |
| Quench++ | Research tooling | Extends Indic reasoning benchmarks with injected bias and structured cause and effect chains in Boolean logic, so trustworthiness and reasoning can be measured repeatably. | Hugging Face · PyTorch · LoRA · QLoRA |
| DreamDOT | Distributed backend | Creator platform with subscriptions and pay per view. A set of GraphQL services answer queries that span services without firing a cascade of extra requests; I set the service boundaries and schema contracts so teams could work in parallel on a shared graph. | GraphQL · Node.js · Microservices · MongoDB |
| Hostiggo | Production SaaS | Idea to a product running on AWS: React front end, full backend, database, and servers, with CI and monitoring from the first week so releases stayed calm as usage grew. | Node.js · React · PostgreSQL · AWS |
| Flourish | Agentic service | Plant care assistant. The FastAPI backend runs sequenced LLM tool calls against external APIs with retries and graceful fallbacks; Firebase handles auth and storage. | FastAPI · LangChain · Firebase · Docker |
| Area | |
|---|---|
| Languages | Python · TypeScript · JavaScript · C++ · SQL · C |
| Backend and APIs | Node.js · Express · FastAPI · REST and GraphQL · GraphQL federation · WebSockets · retries and rate limiting |
| Frontend | React · Next.js · Tailwind CSS · dashboards · live interfaces over WebSockets |
| Infrastructure and delivery | AWS · Docker · GitHub Actions · Linux · Vercel · Pytest · structured logging · static analysis |
| Data and distributed systems | PostgreSQL · MongoDB · Prisma · Pinecone · schema design · indexing · PySpark · Airflow · ETL |
| Machine learning and research | PyTorch · Hugging Face · LoRA and QLoRA · LangChain · RAG · CLIP and Whisper · Google Earth Engine |
First author on a publication currently under review. Title, venue, and co-authors to follow once the review completes.
The work is an evaluation harness that scores model output against several notions of alignment and flags on its own when quality drops; the checking step cut fabricated answers by roughly 20 percent. Harness and benchmarks released as open source.
Open to SDE, MLE, Data Scientist, and Data Engineer roles and internships. Email is the quickest way to reach me: a hard systems problem, a project worth doing, or a good research lead all welcome.
Delhi Metro. Thinking visually, away from the editor.