Skip to content
View gee-46's full-sized avatar
🎯
Focusing
🎯
Focusing

Block or report gee-46

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
gee-46/README.md
Typing SVG

   


🧠 Who I Am

Pre Final-year AI & Data Science student at SGBIT, Belagavi — building systems that make LLMs actually useful: grounded, retrieval-aware, and production-ready.

I don't just use frameworks — I build under them to understand every layer. Currently focused on RAG systems, Computer Vision pipelines, and getting them to production with FastAPI.

Looking for: Remote roles as AI Engineer · ML Engineer · AI Research Intern


🔨 What I'm Building Right Now

Modular RAG System — no LangChain, built from scratch

My main focus. I want to know exactly what's happening at every step — so I skipped the abstractions.

User Query
    ↓
Sentence Embeddings  ←  SentenceTransformers
    ↓
FAISS Vector Search  →  Top-K Relevant Chunks
    ↓
Context Construction
    ↓
Local LLM (Ollama)  ←→  OpenAI  [pluggable backend]
    ↓
Grounded Answer — hallucination significantly reduced

Currently working on:

  • Chunking strategies — sliding window and semantic-aware splitting
  • Retrieval quality benchmarking across chunk sizes
  • FastAPI deployment layer

Why from scratch? I wanted to understand what LangChain actually does. Now I do.


🚀 Featured Projects

🏙️ SlumSafe AI



AI-driven urban safety assessment that identifies risk zones in informal settlements using computer vision + geospatial data. Built end-to-end in 24 hours at Presidency University's International Tech Fest.



AI-powered smart harbor coordination platform — vessel scheduling, anomaly detection, port operations optimization. Built with Team Neural Capitals at BIT Mangalore.



Ask natural language questions about disaster imagery and get answers grounded in what the model sees — not what it was trained to assume. Built on BLIP with a modular model-swap architecture.



Control system volume with hand gestures — no keyboard, no mouse. Real-time hand landmark tracking via MediaPipe mapped to system audio using PyCAW. Works out of the box on CPU.


🛠️ Tech Stack


Area Tools
AI / ML PyTorch · TensorFlow · scikit-learn · SentenceTransformers · HuggingFace Transformers
Retrieval & Search FAISS · Semantic Chunking · Embedding Pipelines · Ollama
Computer Vision OpenCV · MediaPipe · BLIP
Backend FastAPI · REST APIs
Data Pandas · NumPy · Matplotlib
Languages Python · SQL

🏅 Recognition

Event Result
🥉 INNOVATEx 4.0 — International Tech Fest, Presidency University Bangalore 4th Place / 100+ teams
🏅 BIT Techverse'26 — 24hr Hackathon, BIT Mangalore Top 8 / 140+ teams

🎖️ Certifications & Badges

        

 

📊 GitHub Activity

 

📈 Contribution Graph


🐍 Contribution Activity


🏆 GitHub Trophies


📬 Let's Talk

I'm actively looking for remote roles — AI Engineer, ML Engineer, Research Intern. If you're building something where understanding the system matters, let's connect.


I debug for fun. That's either a green flag or a red flag depending on your team culture.

Pinned Loading

  1. SlumSafe-AI SlumSafe-AI Public

    Built for InnovateX 4.0, SlumSafe AI secured 4th place out of 100 teams at the International Tech Fest, standing among the top 10 finalists. A 24-hour sprint that turned ideas into impact and effor…

    Python 3

  2. NEURAL-CAPITALS-HarborPulse NEURAL-CAPITALS-HarborPulse Public

    🚢 HarborPulse is an AI-powered smart harbor coordination platform built by Team Neural Capitals at BIT Techverse’26, securing Top 8 among 140+ teams.

    JavaScript 2

  3. -Disaster-VQA-Response-System -Disaster-VQA-Response-System Public

    "An adaptable, AI-powered Disaster VQA system that performs real-time visual assessment of disaster imagery. By defaulting to the highly efficient BLIP model for rapid inference—while maintaining f…

    CSS 9

  4. gesture-volume-control gesture-volume-control Public

    Real-time gesture-based microphone volume control using MediaPipe, OpenCV & PyCAW — enabling touchless human-computer interaction.

    Python 9

  5. mannmitra-2.0 mannmitra-2.0 Public

    TypeScript 9 1