From debugging 5G radio stacks at Mavenir to building production ML pipelines at UMD — I'm drawn to problems where models meet the real world.
I'm currently an MS student in Applied Machine Learning at the University of Maryland, College Park — but my path here wasn't a straight line.
I spent two years at Mavenir Systems as a 5G systems engineer, doing drive tests, analyzing Grafana/Prometheus logs, and debugging O-RAN deployments in the field. That hands-on experience with real-world data — messy, sparse, domain-specific — shaped how I think about ML now. I don't just want to run models; I want to understand why they work and where they fail.
My sweet spot is the intersection of LLMs, multimodal AI, and real-world deployment — building systems where a model doesn't just predict but actually helps someone make a decision.
When I'm not in the ML weeds, I published an IEEE paper on transfer learning for lung cancer detection, won a hackathon with a computer vision recovery app, and built a privacy-first local RAG system that runs entirely on-device.
MS Applied Machine Learning @ UMD College Park (Sep 2025 – Present)
Privacy-first RAG systems · Applied deep learning · LLM pipelines
LLMs & NLP
Prompt engineering · NER & information extraction · Multimodal reasoning · RAG pipelines · Structured LLM outputs
ML / DL
Computer vision (VGG19, MediaPipe) · Transfer learning · Anomaly detection · Bayesian models · Causal inference
Backend & Data
FastAPI · Celery · Redis · PostgreSQL + pgvector · SQLAlchemy · Alembic
|
FastAPI · pgvector · Ollama · React Privacy-first document Q&A — upload PDFs, embed with |
XGBoost · LightGBM · Random Forest · PCA End-to-end ML pipeline on real O-RAN drive test data. Ensemble model (XGB + LGB + RF) achieves R² = 0.96 predicting downlink throughput from radio signal metrics. Designed a 3GPP-compliant signal imputation algorithm to reconstruct 40% missing SCell measurements. |
|
FastAPI · React · PostgreSQL · MongoDB · WebSockets Real-time SOS alert platform for veterans — geospatial buddy matching (haversine), dual AI provider (Gemini/Ollama), ElevenLabs STT, and hard-wired crisis interception routing to 988 Lifeline before any AI call is made. |
React 19 · Express 5 · PostgreSQL · RTK Query Multi-university campus vending machine platform with Leaflet campus maps, QR payment simulation, and sustainability tracking. Full monorepo with admin dashboard, analytics, and real-time inventory management. |
Cross-Organ Bridge Transfer Learning for Lung Cancer Detection IEEE R10-HTC 2023 · DOI: 10.1109/R10-HTC57504.2023.10461796
Proposed modality-bridge transfer learning — training VGG19 on kidney CT scans as an intermediate domain, then transferring to lung cancer classification. The same-CT-modality bridge improved accuracy from 90% → 93% while cross-modality bridging (USG → CT) degraded to 75%, validating that modality alignment is the key factor.
2025 Hackathon Winner — Best Startup Idea (CV-based postpartum recovery app)
2025 – Now MS Applied Machine Learning · University of Maryland, College Park
2024 – 2025 Member of Technical Staff-I · Mavenir Systems (5G O-RAN, ML on telco data)
2023 – 2024 Graduate Engineer · Mavenir Systems (5G SA system testing, Dish Wireless)
Jan–Jun 2023 Graduate Engineer Intern · Mavenir Systems
2023 IEEE Publication · Cross-Organ Bridge Transfer Learning
2019 – 2023 B.Tech ECE · Amrita School of Engineering, Coimbatore
Always up for a conversation about LLMs, applied AI, or why 5G networks are actually fascinating.