Skip to content
View sumukh30's full-sized avatar
🏃‍♂️
Progress over speed
🏃‍♂️
Progress over speed
4 followers · 8 following

Achievements

Achievement: YOLOAchievement: Pull SharkAchievement: Quickdraw

Achievements

Achievement: YOLOAchievement: Pull SharkAchievement: Quickdraw

Highlights

  • Pro

Block or report sumukh30

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.

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

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

Report abuse
sumukh30/README.md

LinkedIn GitHub Email UIC

👋 About Me

sumukh = {
    "currently"  : "MS Computer Science @ University of Illinois Chicago (GPA: 4.0)",
    "background" : "2+ years Full-Stack Engineering @ Oracle",
    "focus"      : ["LLM Fine-Tuning", "Agentic AI", "Computer Vision", "RAG Systems"],
    "looking_for": "AI/ML Engineering | Full-Stack Engineering roles",
    "location"   : "Chicago, IL"
}

🔩 I bring production engineering discipline to AI/ML — having built and scaled enterprise systems before transitioning to building production-grade AI systems.
⚡ I don't just train models. I ship them.

🚀 Featured Projects

🤖 DataTalk — NL-to-SQL BI Agent

Google Cloud Rapid Agent Hackathon 2026

A multi-step agentic system that lets non-technical users query business data in plain English — no SQL required. Built at the "Google Cloud Rapid Agent Hackathon 2026".

7-step reasoning pipeline:
Schema Introspection → 5 SQL Candidates → BigQuery Execution → Quality Ranking → Human Approval → Insight Generation → History Logging

Stack: Google ADK · Gemini 2.5 Flash · Vertex AI · BigQuery · Fivetran MCP · Firestore · Flask · Cloud Run

🔬 Parameter-Efficient Text-to-SQL

LLM Fine-Tuning Research

Fine-tuned Phi-3.5-mini (3.8B) using LoRA on the Spider benchmark — matching models 4× larger.

Results:

  • 📈 52.5% execution accuracy
  • 🚀 10× exact match improvement (3.5% → 37.5%)
  • ⚡ <1% of parameters updated via LoRA

Stack: PyTorch · HuggingFace · LoRA/PEFT · Spider Benchmark

🎯 Real-Time Prompt-to-Mask Video Analytics

Computer Vision Pipeline

End-to-end drone-view object detection and segmentation — controlled entirely by natural language prompts.

Results:

  • 📊 +177% mAP improvement
  • 🔍 +219% small-object mAP
  • 🎭 SAM-2 mask IoU: 0.8085
  • ⚡ Deployed at 36 FPS in real-time

Stack: YOLO-World · SAM-2 · CLIP · Streamlit · VisDrone

🌐 Oracle Enterprise Portal

Production — 13M+ Users

Full-stack web portal serving 13M+ users at Oracle — REST APIs, frontend systems, B2B data pipelines, and ML-powered sales lead scoring.

Highlights:

  • 🏗️ REST API latency reduced 15–25%
  • ⚙️ Groovy automation: 10–20s → instant
  • 🤖 ML classification for sales lead conversion probability
  • 📉 95% manual effort eliminated

Stack: HTML · CSS · JavaScript · REST APIs · Groovy · Oracle B2B Cloud

🛠️ Tech Stack

Languages
Python JavaScript SQL HTML5 CSS3

AI / ML
PyTorch HuggingFace Scikit--learn OpenCV

Cloud & MLOps
Google Cloud Vertex AI Azure BigQuery

Web & Backend
Flask React Node.js Git

📊 GitHub Stats

🎓 Education & Experience
🎓 MS Computer Science — University of Illinois Chicago Aug 2025 – May 2027 · GPA: 4.0 / 4.0
🎓 BE Computer Science — Visvesvaraya Technological University Aug 2018 – Jul 2022 · GPA: 3.36 / 4.0
💼 Senior Cloud Analyst — Oracle Sep 2024 – Oct 2024
💼 Cloud Analyst — Oracle Aug 2022 – Aug 2024
💼 Software Development Intern — Phamax Dec 2021 – Mar 2022

🧠 What I'm Working On

  • ⚡ Learning by developing scalable, production-ready AI applications using modern AI, cloud, and software engineering tools to solve real-world problems
  • 🎯 Targeting AI/ML Engineer and Full-Stack Engineer roles

Building things that matter - one commit at a time.

Pinned Loading

  1. datatalk-bi-agent datatalk-bi-agent Public

    Natural language BI agent powered by Gemini, Fivetran MCP, and BigQuery. Ask business questions in plain English — with human-in-the-loop SQL approval

    HTML

  2. schema-aware-text2sql schema-aware-text2sql Public

    Parameter-efficient Text-to-SQL pipeline using LoRA on Phi-3.5-mini, with schema validity checking and execution-guided reranking. Evaluated on Spider benchmark

    Jupyter Notebook

  3. ov-prompt-to-mask ov-prompt-to-mask Public

    Open-vocabulary drone-view object detection, pixel segmentation, and temporal tracking — type a text prompt, get tracked masks across every video frame. YOLO-World + SAM-2, fine-tuned on VisDrone (…

    Jupyter Notebook 1

  4. linkedin-post-automation-n8n linkedin-post-automation-n8n Public

    Local n8n workflow for AI-assisted LinkedIn post drafting, approval, and publishing using RSS, Google Sheets, and Gemini.

  5. budgetWise budgetWise Public

    Forked from imandeol/budgetWise

    Project for CS 480 : Database Systems

    TypeScript

  6. Wildlife-Database Wildlife-Database Public

    DBMS project on Animal Wildlife database system

    Hack