Los Angeles, California
Khushi Prashant
Mehta.
Data Scientist·Data Analyst·AI/ML Engineer
I'm a Master's student in Computer Science at USC, building production ML and GenAI systems — from Text-to-SQL agents to satellite-imagery classifiers — after cutting my teeth on AI research at ISRO and AI-driven product design at Hashtechy.
▶ About
I'm currently pursuing my Master's in Computer Science at USC (expected Dec 2026). Most recently, I was a Machine Learning Intern at ASML, where I designed and deployed Pulse Agent — a full-stack GenAI app that lets business users ask questions in natural language, converts them to SQL over structured operations data, and surfaces results in Power BI. I fine-tuned an LLM on domain-specific SQL, architected a RAG pipeline over fragmented process documentation, and built time-series forecasting across 24,300+ shipment records that surfaced a 28% quarter-over-quarter cost increase for logistics leadership.
I currently TA "AI for Data Science" at USC Marshall, reviewing and debugging ML codebases for 40+ students and mentoring predictive-analytics projects end to end. Before USC, I was an AI UI Front-End Designer at Hashtechy, building AI-driven interfaces for the "Currently" social app — which grew to 500K+ downloads and a Shark Tank India feature — and running 100+ A/B tests. And before that, an AI/ML Research Intern at ISRO's Space Application Center, training U-Net models and benchmarking classical ML against deep learning for satellite-imagery land-cover classification.
⁕ What I Do
Areas of expertise
Six areas where modeling, data, and interface design meet.
01 / ML
Machine Learning & Deep Learning
PyTorch, TensorFlow, Keras, Scikit-learn, CNNs, NLP, computer vision, time-series forecasting, recommender systems.
02 / AI
Generative AI & Agentic Systems
LLM fine-tuning, RAG, LangChain, LangGraph, LlamaIndex, CrewAI, Model Context Protocol, vector databases (FAISS, ChromaDB).
03 / DS
Data Science & Analytics
Statistics & probability, A/B testing, experiment design, Pandas & NumPy workflows at scale.
04 / OPS
Data Engineering & MLOps
ETL pipelines, FastAPI, MLflow, Docker, CI/CD, AWS (EC2, S3, SageMaker), Azure ML, Google Vertex AI.
05 / BI
BI & Dashboarding
Power BI, interactive dashboards, ArcGIS & ArcPy for geospatial data.
06 / UI
AI-Driven UI / Front-End Design
Turning model outputs into personalized, production UI components — Figma, Adobe XD.
▶ Toolkit
Skills & technologies
Languages
Machine Learning & Deep Learning
Generative AI & LLMs
Data Engineering & MLOps
Frameworks, BI & Design
⁕ Career
Experience & education
-
May 2026 – Aug 2026
Machine Learning Intern, Pulse Agent
ASML · San Diego, CA
- Designed and deployed Pulse Agent — a full-stack GenAI app converting natural-language questions to SQL over structured ops data, surfacing results in Power BI.
- Fine-tuned an LLM on domain-specific SQL; architected a RAG pipeline consolidating fragmented process documentation into one knowledge base.
- Built time-series forecasting across 24,300+ shipment records, surfacing a 28% Q4-over-Q1 cost increase for logistics and finance leadership.
-
May 2025 – Present
Teaching Assistant, AI for Data Science
USC Marshall School of Business · Los Angeles, CA
- Reviews and debugs Python ML codebases for 40+ students, diagnosing logic defects and performance bottlenecks.
- Supports end-to-end ML/NLP pipelines in TensorFlow and scikit-learn, from preprocessing to evaluation.
- Mentors predictive-analytics projects from problem definition to working solution.
-
Sep 2023 – Dec 2024
AI UI Front-End Designer
Hashtechy · Ahmedabad, India
- Designed AI-driven front-end interfaces for the "Currently" social platform, integrating ML outputs into personalized UI — growth to 500K+ downloads and a Shark Tank India feature.
- Ran 100+ A/B tests using SQL and Python to improve retention; built executive-facing data visualizations.
-
Jan 2023 – Jul 2023
AI/ML Research Intern
Space Application Center, ISRO · Ahmedabad, India
- Benchmarked classical ML (k-means, Random Forest) against deep learning for land-cover classification using ArcGIS/ArcPy.
- Trained a U-Net CNN for glacier segmentation, improving classification accuracy by 20–30% across 1,000+ Sentinel-2 images.
↗ Selected Work
Projects
A few things I've built — from agentic tooling to full-stack GenAI products.
MCP-Powered AI Assistant
An AI assistant on Anthropic's Model Context Protocol with a decoupled tool server (web search, file I/O, REST API) and LLM client — Claude/Groq provider swaps with no code changes, sandboxed tool execution.
Loan Approval Prediction
Logistic regression model, 81% accuracy on 600+ applications, deployed via Streamlit with OpenAI-generated plain-language explanations — cut decision time ~40%.
AI-Powered Travel Planning Assistant
Full-stack GenAI assistant generating personalized itineraries from natural-language queries, with multi-turn memory for follow-up requests.
Glacier Classification with Satellite Imagery
U-Net CNN for semantic segmentation of Sentinel-2 imagery, benchmarked against Random Forest and k-means — improving classification accuracy 20–30%.
Flavor Finder
Flask + ML food recommendation API trained on Indian cuisine data, paired with LLM-powered suggestions via OpenRouter.
5×5 Go Game AI
AI-powered 5×5 Go game using Minimax search with Alpha-Beta pruning for efficient move evaluation.
↗ Get In Touch
Let's build something worth using.
Open to data science, analytics, and AI/ML roles — or just a conversation about GenAI, modeling, or design. I usually reply within a day or two.