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 DetectionIEEE 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.
