Charvi Shukla 👩💻
👋🏽 Hi! I am a fourth year undergraduate student at UC San Diego, pursuing a dual degree in Mathematics-Computer Science and Cognitive Science with a specialization in Machine Learning and Neural Computation.
My research interests lie in medical and autonomous driving applications of Computer Vision. At UCSD, I have worked as an instructional assistant for the Computer Science and Engineering Department (CSE 30, CSE 105), and as a Supplemental Instruction leader for the Mathematics department (MATH 18) through the Academic Achievement Hub. I also work as an undergraduate student researcher at the Advanced Robotics and Controls Lab at UCSD.
🤗 Experience
🤖 Advanced Robotics and Controls Lab @ UCSD
LLM fine-tuning
- Processed a large medical-assessments dataset and used LoRA finetuning to fine tune Meta Llama 2 and Google Gemma models with 7 billion parameters
- Used ROUGE, BERT, and Bleu Scores each to evaluate and compare the model's performance against human evaluators
- Deployed CUDA based Docker containers to efficiently train models using the PyTorch DataParallel class, leveraging 4 GPUs in parallel
SuPerPM, A Surgical Perception Framework
- Fine tuned Lepard, a point cloud matching model for rigid and deformable scenes, over a physically constrained simulation dataset
- Implemented a custom visualization algorithm using Open3D to visualize 3D point clouds and their correspondences
⚖️ WiseCounsel.ai
- Developed core negotiation engine CLI for WiseCounsel's end-to-end deal-making platform, enabling structured bi-directional communication between clients and counterparties
- Enhanced CLI functionality to parse and reconstruct highly structured legal agreements, orchestrating multi-clause negotiations by managing LLM inputs/outputs and agreement lifecycle
🔶 Association of Computing Machinery at UC San Diego
Hack Technical Events Director
- Developed a 6-week workshop series (ACM Hack School) on MERN-Stack app development. Worked in a team of 5, to create an online textbook explaining core concepts such as HTML/CSS, React.js, MongoDB, Express.js and Node.js, and created a web based type-racer game for teaching.
- Link to the Hack School Textbook
- Link to the Type Racer Github
- Held a two-part workshop series on Intro to Building Chrome Extensions. Leveraged Gemini-Pro API from Google AI studio to create a transcript summarizer for recorded lectures
- Link to the Transcript Summarizer Github
ACM Hack Project Mentor
- Mentored a team of 6 developers to develop a full-stack web application using the MERN stack in 10 weeks
- Held weekly meetings and guided students on the design and implementation of the web-application. Presented the final product during the Quaterly ACM Projects Demo.
📚 Tutoring and Supplemental Instruction
Academic Achievement Hub, MATH 18: Linear Algebra
- Supplemental Instruction leader in Spring 2022, Winter 2022 and Fall 2023
- Collaborated with Teaching Assistants to conduct 5 weekly discussion sessions for 20-30 undergraduates each, reinforcing Linear Algebra lecture content among students
- Orchestrated supplemental instruction in groups of 10 students, engaging students in problem-solving and doubt clarifications
Computer Science and Engineering Tutor
- Spring 2024: CSE 30, Computer Organization and Systems Programming
- Conducted 4.5 hours of weekly tutoring for students, debugging C and ARM Assembly code and re-iterating core class concepts
- Fall 2023: CSE 105, Theory of Computation
- Tasks include holding office hours, proctoring and grading assignments.
🚀 Projects
- Worked in a team of 2 to develop an IoT device for individuals with hearing disabilities
- Trained a LSTM Neural network using PyTorch and MediaPipe with manually collected data to recognize hand gestures and control smart-home devices
- Developed a voice-activated recipe generation application using JavaFX and Gradle, integrating Whisper API for audio input recognition of pantry items and meal preferences
- Engineered a RESTful backend in Java to manage application data flows and API integration with GPT-3.5 to generate tailored recipe suggestions based on user inputs and DALL-E API for image generation
- Ensured robust application performance through comprehensive testing strategies, including unit testing, behavior-driven development (BDE) testing, and end-to-end (E2E) testing.