Software engineer with 4 years at BlackRock, building distributed backend systems, regulatory technology, and agentic AI infrastructure on the Aladdin platform.

Now pursuing my MS in Computer Science (AI) at the University of Southern California, and working as a Research Assistant at the USC Stevens Neuroimaging and Informatics Institute.

I like problems that sit at the seam between large-scale systems and machine learning — LLM tooling that has to survive real production traffic, not just a demo.

Right now

• 🔬 Researching contrastive learning for MRI-based neurodegenerative disease classification
• 🌱 Contributing to open source: Capital One's DataComPy, Morgan Stanley's url-detector, ServiceNow's SlackerBot
• 🏁 Competing in the OpenCV AI Competition 2026, powered by AWS
• 🛠️ Building an adaptive distributed rate limiter and practicing DSA in Java on NeetCode

Sep 2026 – Present · Los Angeles, CA

Research under Dr. Paul Thompson on metadata-informed contrastive learning for MRI-based neurodegenerative disease classification, building on the y-Aware InfoNCE framework (MICCAI 2021).

Associate, SDE 2 (Jan 2025 – Jul 2026) · Analyst, SDE 1 (Jul 2022 – Dec 2024) · Gurugram, India

• Agentic AI: Built an MCP Server for API infrastructure with LangChain and LLMs, turning 80+ natural-language prompts a day into executable, context-aware API calls. Shipped an end-to-end RAG pipeline (LangChain + ChromaDB + Azure DevOps REST APIs) that cut manual release-note effort by 70%.
• Identity & access: Core developer on user management APIs serving 1M+ users across 200+ clients, with gRPC User Services and SCIM-compliant REST APIs for identity and access governance. Launched an IdentityProvider Management Portal that cut SSO client onboarding from 1 day to 30 minutes.
• Regulatory technology: Designed Java/Spring Boot microservices powering 6+ global OTC derivative regulations, and orchestrated Airflow pipelines for regulatory reporting across 6+ jurisdictions. Built a Reporting Validation Platform (Template Method pattern) that cut processing time by 80%.
• Reliability: Rolled out Grafana and Prometheus observability across 20+ services, and automated memory profiling and leak detection in the Jenkins/Docker CI/CD pipeline, cutting production issues by 60%.
• Mentorship: Led onboarding and mentorship for 150+ new-hire engineers across 2 batches, running training on distributed systems, ML pipelines, and platform architecture.

Tech: Java, Spring Boot, Python, LangChain, ChromaDB, gRPC, Protocol Buffers, Airflow, Grafana, Prometheus, Jenkins, Docker, Azure

Jan 2022 – Jun 2022 · Gurugram, India

Built a Data Harness utility for Aladdin applications and orchestrated workflow pipelines across 100+ clients. Automated workflows that prevent first-click issues, improving client-onboarding reliability by 50%. Received a return offer for full-time.

Project	Contribution	Status
Capital One · DataComPy	#564 — Added a DataComPy agent skill: SKILL.md, backend reference, and a v0 → v1 migration guide
Morgan Stanley · url-detector	#199 — Granular CLI exit codes, so CI can tell "URLs found" apart from config and file errors
	#198 — --ignore-domains-file flag to load ignore-domain patterns from a file
ServiceNow · SlackerBot	#502 — Fixed !convert by replacing a dead exchange-rate API
	#503 — New !units parser for length conversions

USC's startup launch sprint · Sep 2026 · Python · Go · Java · Docker · Google Gemini · AWS

A secure, centralized library for the small tools employees build to automate their company's SOPs. Publish a tool from a GitHub link, share it with one person or the whole company, and colleagues find it by describing the task in plain language, then run it in the browser.

• One manifest per tool: a tool.yaml declares the Dockerfile, command, inputs, outputs, resources, and sharing scope, and Gerson turns it into a run form. Example tools in Python (data-comparator, csv-profiler), Java (compliance-checker), and Go (trade-pricer) show the format is language-agnostic.
• Chat-based discovery: Gemini reads the manifests and ranks the tools that fit a described task, falling back to plain catalog ranking when the model is unreachable.
• Pay per run: every run gets its own container. Three runs cost $0.000012, against $36 a month to keep the same container running on Fargate or Azure.
• Customer discovery: 20 interviews at 12 companies; 15 of 20 people had already automated a task their company documents as a formal procedure.

🎥 Demo video · 👥 Six-person team

Building a vision entry on OpenCV 5 and AWS; final submissions are due Oct 26, 2026.

Java 21 · Spring Boot 4 · PostgreSQL · Redis · WebSocket · Flyway · Testcontainers

Matches players for multiplayer games in real time. Queues live in Redis sorted sets keyed by Elo, one per region and game mode; a scheduled worker groups players within an Elo threshold every second, persists the match to PostgreSQL, and pushes a "match found" event over WebSocket. Candidate matches are ranked by a pluggable match-quality scorer (Elo spread, team balance, win rate, quit rate) that is rule-based today and built to swap in an ML model.

TypeScript · Express · Redis + Lua · Docker Compose · nginx · k6

A rate limiter that stays correct across horizontally scaled app instances: Redis holds the shared state and every check runs as one atomic Lua script, so concurrent requests can't race on the same counter. Covers fixed window, sliding window log, and token bucket, plus an adaptive mode that slows a client's refill rate when its traffic drifts from its own baseline (EWMA + z-score).

Python · PyTorch · FAISS · 2025

Vision-language pipeline for retrieving MRI scans and answering questions about them, using FAISS vector search over image-text embeddings. Reached 0.89 AUC.

Flutter · Spring Boot · PostgreSQL · Redis · AWS S3

A short-video platform for technical interview prep. JWT-authenticated Spring Boot APIs with Redis-backed feed caching, and a Flutter client with a rule-based recommendation engine that personalizes feeds by followed creators, watch history, and popularity.

Logistic regression from scratch in PyTorch · Twitter sentiment analysis · Stock trend prediction · Diabetes prediction with KNN

📘 IoT-Based Women Safety Gadgets: Vision, Architecture, and Design Trends
Computers, Materials & Continua (2021) — 500+ views and downloads. Led a team of 4 building a safety device with pepper spray, geotagged location tracking, and emergency SOS alerts.

📗 Visual Speech Recognition for Speech-Impaired Individuals
Wiley (2021) — Real-time lip-reading system: a spatio-temporal encoder and descent-based decoder that generate speech and text transcriptions for speech-impaired users.

📝 Detecting Out-of-Memory Issues in Spring Boot Servers
BlackRock internal tech platform (2024) — Using SpringBootTest under restrictive environment configs to catch OOM failures early in the SDLC.

🔗 Full list on Google Scholar

• University of Southern California — MS, Computer Science (Artificial Intelligence) · Aug 2026 – May 2028
• Thapar Institute of Engineering and Technology — BE, Computer Engineering · CGPA 8.88 · Aug 2018 – May 2022

• AI & LLMs: LangChain · MCP · RAG · agentic AI frameworks · prompt engineering · HuggingFace · FAISS · ChromaDB
• Also: gRPC · Protocol Buffers · Airflow · Flyway · Testcontainers · Selenium · JUnit

• 🦁 Treasurer, Leo Club Gurugram Cybercity — youth leadership and community service under Lions Club International
• 🤝 Volunteer, BlackRock Gives — 40+ hours with non-profit partners
• 📜 Certifications: Azure Fundamentals (AZ-900) · Machine Learning (Stanford, Coursera) · DeepLearning.AI Specialization · ServiceNow Micro-Certification · Self-Driving Cars (Univ. of Toronto)