MS Computer Science • AI/ML + Full-Stack
Thesis focused on anomaly detection for deepfake facial images. I build ML pipelines, dashboards, and production-ready web apps — with strong experimentation and engineering.
I’m Mili Patel — building AI systems that don’t break when the world changes, and shipping full-stack products that make those systems usable.
My thesis focuses on detecting deepfake facial images under real-world distribution shifts (new manipulations, unseen datasets, changing lighting, compression artifacts, and model drift). That means I care about robustness, calibration, and decision boundaries — not just accuracy on a single benchmark.
Alongside research, I build production-grade systems: clean UI, fast APIs, evaluation dashboards, and reproducible pipelines. I like work where machine learning meets engineering — the part where results become tools.
Training on clean splits, validating with controlled OOD sets, and testing against unseen manipulation methods — with score analysis + dashboarding for model behavior.
If it can’t be evaluated, monitored, and explained — it isn’t done.
Academic foundation in AI, machine learning, and systems.
Full-stack builds, ML research, dashboards, and edge projects — all in one place.
Research-driven anomaly/OOD detection for deepfake facial images under distribution shift — beyond closed-set classification.
Real-time retail dashboard with alerts, email notifications, and exports.
Large-scale crime analytics with dashboards for hotspots, trends, and outcomes.
Raspberry Pi + OpenCV pipeline for real-time face recognition and attendance automation.
YOLOv3 inference optimization on PYNQ FPGA for lower latency object detection.
ML model to flag likely journal rejections + Streamlit prototype for formatting issues.
Captioning tool for accessibility with live speech-to-text + multilingual translation.
A curated view of the tools I use to build ML systems and full-stack products.
A timeline of my roles — teaching, internships, leadership, and hands-on engineering work.
Peer-reviewed work and research contributions — plus my ongoing thesis work in deepfake anomaly detection.
Detecting deepfake facial images under distribution shift using anomaly/OOD signals — with robust evaluation and behavior analysis.
Archives of Computational Methods in Engineering (Springer)
A curated collection of recognitions, hackathons, courses, and achievements.
Want to collaborate, discuss research, or talk full-stack? Send a message — I reply fast.