https://github.com/stevehoober254/ai-portfolio
🧠LLM pipelines, model training notebooks, and real-world AI/ML deployments with testing
https://github.com/stevehoober254/ai-portfolio
ai deep-learning huggingface llm machine-learning mlops model-training notebooks pytorch transformers
Last synced: 9 months ago
JSON representation
🧠LLM pipelines, model training notebooks, and real-world AI/ML deployments with testing
- Host: GitHub
- URL: https://github.com/stevehoober254/ai-portfolio
- Owner: stevehoober254
- Created: 2025-04-10T13:48:57.000Z (over 1 year ago)
- Default Branch: main
- Last Pushed: 2025-04-10T19:52:36.000Z (over 1 year ago)
- Last Synced: 2025-04-13T11:14:32.242Z (over 1 year ago)
- Topics: ai, deep-learning, huggingface, llm, machine-learning, mlops, model-training, notebooks, pytorch, transformers
- Homepage:
- Size: 6.84 KB
- Stars: 1
- Watchers: 1
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- Funding: .github/FUNDING.yml
Awesome Lists containing this project
README
# 🤖 AI Engineer Portfolio
This portfolio features advanced AI/ML projects with production-ready pipelines. It covers everything from LLM integration and model fine-tuning to deploying ML services and validating notebooks with CI workflows.
## 🧠Highlights
- LLMs with Hugging Face Transformers
- Model training and evaluation notebooks
- MLOps-style workflow testing (nbmake, CI)
- Checkpointing and model lifecycle
- AI deployments with FastAPI & Docker
## Project List
## 1. Multilingual Voice Agent for Public Services
### Problem
Citizens struggle to access government info in native languages via digital channels.
### Solution
Build a voice assistant that:
- Converts speech to text using **Whisper**
- Translates using **NLLB (No Language Left Behind)**
- Queries a knowledge base using **RAG with LlamaIndex**
- Speaks answers via **TTS (Bark or Coqui)**
- Deployable via **Twilio Voice** or WhatsApp
### Goals
- Serve local languages like Swahili, Luo, Kikuyu, Somali
- Provide answers about IDs, licenses, schools, etc.
- Log anonymized usage metrics
---
## 2. African News Summarizer and Bias Detector
### Problem
News often contains bias and lacks multilingual accessibility.
### Solution
Train a summarizer + bias detector:
- **Fine-tune T5 or Falcon** on local news corpora
- Build React frontend to enter URLs and summarize
- Use **Hugging Face Transformers + Langchain**
### Goals
- Summarize articles in 3 languages
- Classify for bias type: political, regional, tone
- Suggest counterpoints and sources
---
## 3. Medical Imaging Assistant for Understaffed Clinics
### Problem
Clinics lack radiologists to interpret X-rays and CT scans.
### Solution
Build a diagnostic model using:
- **YOLOv8 + fastai** for object detection
- Train on open datasets (NIH ChestX-ray14, VinDr)
- Deploy via **Streamlit** or **Gradio** interface
### Goals
- Identify common findings (TB, pneumonia, fractures)
- Allow image uploads from mobile
- Generate PDF reports