Ashmit Sethi
Applied Math & Data Science Student
Passionate about machine learning, AI research, and building innovative solutions. Currently pursuing my B.A. at UC Berkeley with a focus on data science and applied mathematics.
About Me
I'm a dedicated student at UC Berkeley pursuing a Bachelor's of Arts in Data Science and Applied Math. My passion lies in machine learning research, GenAI development, and creating impactful solutions that bridge the gap between theoretical knowledge and real-world applications.
With experience spanning software engineering, academic research, and practical implementation, I've worked on projects ranging from RAG pipelines and computer vision for healthcare to fairness-aware machine learning systems.
Technical Skills
Programming Languages
Machine Learning & GenAI
Frontend & Backend
Tools & Cloud
Education
University of California, Berkeley
Expected Graduation: May 2028Bachelor's of Arts in Data Science and Applied Math
Relevant Coursework:
- Data Structures & Algorithms (DSA)
- Object-Oriented Programming (OOP)
- Data Science
- Data Visualization
Work Experience
Software Engineering Intern
Stealth Startup (Healthcare) Sep. 2025 - PresentBerkeley, CA
- Developed internal RAG pipelines integrated with Model Context Protocol (MCP) to improve GenAI-driven medical document retrieval and summarization while maintaining HIPAA compliance.
- Contributed to computer vision workflows for medical imaging by implementing model training, dataset preprocessing, and inference optimization using PyTorch.
- Created data models and Power BI dashboards to visualize LLM and CV system metrics, supporting data-driven product and research decisions.
Machine Learning Research Assistant
Indiana University, Bloomington Aug. 2024 - Nov. 2024San Jose, CA
- Supported LLM research in the medical domain under Professor Da Yan and Ph.D. student Saugat Adhikari by annotating several hundred data samples for training and evaluation tasks.
- Assisted with PyTorch model training workflows, including dataset loading and basic model evaluation for NLP-related tasks.
- Explored prompt-engineering and data-augmentation techniques to enhance LLM fine-tuning efficiency for domain-specific text generation.
Machine Learning Research Assistant
University of Texas, San Antonio May. 2024 - Jul. 2024San Jose, CA
- Conducted research under the CAREAI cohort with Professor Ke Yang and Ph.D. Jason Johnson, focusing on counterfactual generation to improve classifier fairness and enhance model interpretability.
- Collaborated on implementing and testing Python-based ML models, analyzing the fairness impact across multiple datasets using a range of standard fairness metrics.
- Supported benchmarking of fairness algorithms by visualizing bias metrics and comparative results using Python and Matplotlib.
Featured Projects
Multimodal Agentic AI Desktop Operator
Built a multimodal agentic AI system that autonomously controls computer interfaces (mouse, keyboard, and applications) via natural voice or text using LLM-driven reasoning. 1st Place Winner at Cal Hacks 12.0 (ElevenLabs).
- Integrated real-time computer vision and GUI automation to execute multi-application workflows with high precision and full end-to-end autonomy
- LLM-driven reasoning engine enables natural language interaction for complex task automation
- Multimodal input support (voice and text) with seamless interface control capabilities
Computer Vision Gym Tracker
Developed an AI-powered fitness tracker leveraging MediaPipe pose estimation and CNN-based joint analysis for real-time form correction and motion tracking. 1st Place at Tri Valley Hackathon.
- Implemented optimized OpenCV pipelines with efficient frame handling and latency reduction, improving visual feedback responsiveness and accuracy
- CNN-based joint analysis enables precise exercise form evaluation and real-time correction
- Real-time pose estimation with automated form feedback for users
AI Image Authenticity Detector
Trained deep learning models (EfficientNet, ResNet, ViT) to detect AI-generated images with 84%+ accuracy using a robust data augmentation and explainability pipeline.
- Deployed GPU-accelerated Flask and Streamlit apps with Grad-CAM visualization for interpretable AI image verification
- Robust data augmentation pipeline improves model generalization and accuracy
- Real-time inference capabilities with explainability features for transparency
DataPulse - AI-Powered Information Dashboard
A comprehensive real-time dashboard that aggregates crypto, stocks, weather, and news data with intelligent AI agents that analyze market movements and provide insights.
- Built 5 AI agents using BrightData API to automatically research and explain stock price movements
- Real-time data aggregation from multiple APIs with Redis caching for optimal performance
- Responsive React frontend with dark mode and mobile-first design
- FastAPI backend with automated CI/CD pipeline deployed on Render
- Live demo: data-pulse-xvft.vercel.app
Leadership & Impact
Founder - HowToHackathon
May. 2023 - May. 2025- Founded and led 25+ global hackathons and olympiads with 25,000+ participants
- Awarded $2.5M+ in prizes, fostering large-scale technical innovation
- Built partnerships with 10+ industry leaders (Wolfram, Postman, Desmos)
- Connected 100+ judges from companies like Meta, Google, and Nvidia
- Managed technical operations and a 20+ person team
President - DevFinTech
Dec. 2022 - May. 2025- Co-founded and led an organization with 100+ global chapters
- Educated 5,000+ students through hands-on FinTech courses
- Spearheaded events in stock market simulations and cryptocurrency trading
- Expanded access to FinTech and computing education globally
Awards & Recognition
Get In Touch
I'm always interested in discussing new opportunities, research collaborations, or just having a conversation about technology and innovation.