Skip to content
View omavashia2005's full-sized avatar

Highlights

  • Pro

Block or report omavashia2005

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Please don't include any personal information such as legal names or email addresses. Maximum 250 characters, markdown supported. This note will be visible to only you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
omavashia2005/README.md

Hi, I'm Om Avashia 👋

👨‍💻 AI Research Volunteer @ ASU APG
🕙 Prev Software Engineer Intern @ IP Author
🚀 Junior @ Arizona State University
🎓 B.S. in Computer Science
🧠 Backend & AI Engineer | Open Source Contributor


🛠️ Currently Working On

AI Research Volunteer @ ASU APG

  • Developed instrumentation to analyze 10K+ LLM training epochs, enabling research on LLM generalization and grokking
  • Produced 300+ Hessian top eigenvalues, loss landscapes, and weight norms to visualize and quantify experiment results
  • Trained a small LLM on the Sol Supercomputer with NVIDIA L40 GPUs in CUDA-accelerated environments (PyTorch + Mamba), using Matplotlib for data visualization of key metrics collected during experimentation

🕙 Previous Work Experience

✍️ LLM-Powered Patenting @ IP Author

Tech: Python · FastAPI · React · Flask · TipTap · Dify.ai

  • Reduced feature development time by 50% by refactoring full-stack systems (Python/Flask + React), improving the scalability and usability of AI-driven processes, used by clients like Google and Siemens
  • Contributed to Dify’s open source AI workflow platform, enhancing the Python SDK with new imports, documentation, and testing, and collaborating with the team to improve developer experience and workflow reproducibility
  • Worked with a 5-person cross-functional team (tech lead, backend, frontend, prompt engineer, and founder) to plan end-to-end features, conduct code reviews, and align on iterative rollout schedules
  • Followed structured SDLC using Jira to manage milestones, branches, and PR workflows across full-stack deliverables

Open Source Contributions

  • 🛠️ (Open) PR to Kubernetes Python client (kubernetes-client/python#2420): Fixing a bug in RFC3339 timestamp parsing. Added error handling and unit tests to prevent crashes when config datetime strings are invalid.

  • ✅ (Merged) PR to Dify’s Python SDK (langgenius/dify#22476): Exposed missing imports and added documentation for WorkflowClient and KnowledgeBaseClient

  • 📊 Published a musical scale classification dataset on Kaggle with 400+ views and 20+ downloads, built as part of ChromaLite — a PyTorch-based neural network that achieved 80% test accuracy on deduplicated chromagram inputs.


💻 Tech Stack

  • Languages: Python · Java · C++ · JavaScript (React, Next.js) · TypeScript
  • Frameworks & Tools: Spring Boot · FastAPI · Flask · JWT · REST APIs · JPA · Git · Postman · Unit Testing
  • Databases: PostgreSQL · MySQL · Redis · Supabase
  • ML/AI: PyTorch · CNNs · Feature Engineering · Synthetic Data Generation
  • Interests: System Design · Offensive Security · Streamlit Dashboards · Log Analysis

📫 Let's Connect

Pinned Loading

  1. om-portfolio om-portfolio Public

    Portfolio website built with React + Next.js

    JavaScript

  2. expenseTracker expenseTracker Public

    A fullstack app with a Spring Boot backend for tracking expenses, featuring JWT-based authentication, BCrypt password hashing, PostgreSQL integration, and a layered architecture, complete with a St…

    Java

  3. ChromaLite ChromaLite Public

    A simple, custom neural network trained on 1,900+ synthetically generated chromagrams to classify Western musical scales.

    Jupyter Notebook

  4. Rainfall-Prediction-using-ML Rainfall-Prediction-using-ML Public

    This project is a machine learning pipeline built from scratch to predict rainfall.

    Jupyter Notebook