Skip to content

SadiqCodex/GenAI-Learning-Project

BranchesTags

Repository files navigation

GenAI Learning — LangChain & LLM Projects

A hands-on collection of Generative AI projects and experiments built with LangChain, Ollama (local LLMs), Streamlit, and various AI APIs. This repository covers the full spectrum of modern GenAI development — from basic LLM calls and embeddings to multi-agent pipelines, RAG systems, and production-ready AI applications.

Table of Contents

🛠 Tech Stack

Category Tools
LLM Framework LangChain, LangChain-Ollama
Local LLMs Ollama (phi3, nomic-embed-text)
External APIs Tavily Search, Sarvam AI (STT), TMDB, OpenWeatherMap
Vector Stores FAISS, ChromaDB
Embeddings Ollama Embeddings, HuggingFace (all-MiniLM-L6-v2)
Audio/Video Whisper (OpenAI), yt-dlp, pydub, FFmpeg
Web Framework Streamlit, FastAPI
Deployment Render (movie-rec API)
Language Python 3.11+

📁 Project Structure

GenAI Learning/
├── AI-Video-Assistant/          # YouTube/video transcription + RAG chat
│   ├── core/
│   │   ├── extractor.py         # Action items, decisions, questions extraction
│   │   ├── rag_engine.py        # LCEL RAG pipeline over transcript
│   │   ├── summarizer.py        # Map-reduce summarization
│   │   ├── transcriber.py       # Whisper + Sarvam AI (Hinglish support)
│   │   └── vector_store.py      # ChromaDB + HuggingFace embeddings
│   ├── utils/
│   │   └── audio_processor.py   # yt-dlp download, WAV conversion, chunking
│   └── app.py                   # Streamlit UI
│
├── Multi-agent-research-system/ # 4-agent research pipeline
│   ├── agents.py                # Search agent, Reader agent, Writer & Critic chains
│   ├── pipeline.py              # CLI pipeline runner
│   ├── tools.py                 # Tavily web search + BeautifulSoup scraper tools
│   └── app.py                   # Streamlit UI with live pipeline status
│
├── rag/                         # RAG system with PDF ingestion
│   ├── app.py                   # Streamlit UI (upload PDF → ask questions)
│   ├── create_database.py       # PDF → FAISS vector store builder
│   ├── main.py                  # CLI RAG query interface
│   ├── document loaders/        # PDF, web page, text loaders (experiments)
│   ├── retrievers/              # MMR, multi-query, arXiv retriever experiments
│   └── vector store/            # ChromaDB experiments
│
├── movie-rec-ai/                # Movie recommendation system
│   ├── main.py                  # FastAPI backend (TF-IDF + TMDB API)
│   ├── app.py                   # Streamlit frontend
│   └── movies.ipynb             # Data preprocessing & TF-IDF model building
│
├── cinesage/                    # Movie info extractor using structured output
│   ├── core.py                  # CLI version (Pydantic + Ollama)
│   └── UICore.py                # Streamlit UI version
│
├── chatmodels/                  # LLM chat experiments
│   ├── chat.py                  # Basic LLM invocation (Ollama/OpenAI/Groq/Gemini)
│   ├── chatbot.py               # CLI chatbot with personality modes + memory
│   └── UIChatbot.py             # Streamlit chatbot with 5 AI personality modes
│
├── tools/                       # LangChain tools & agents experiments
│   ├── toolcalling.py           # Tool binding + manual tool call loop
│   ├── owntool.py               # Custom @tool decorator example
│   ├── Agents.py                # City agent (weather + news) with human approval
│   ├── newssummarizer.py        # Tavily search → LLM summarization chain
│   ├── sequencerunnable.py      # LCEL RunnableSequence experiments
│   ├── parallelrunnable.py      # LCEL RunnableParallel experiments
│   └── runnablepassthrough.py   # RunnablePassthrough experiments
│
├── embeddingmodels/             # Embedding model experiments
│   ├── embedding.py             # Ollama embeddings (nomic-embed-text)
│   └── huggingfaceembedding.py  # HuggingFace embeddings
│
├── .env                         # API keys (not committed)
├── requirements.txt             # All Python dependencies
└── pyproject.toml               # Project config (uv)

🚀 Projects

1. AI Video Assistant

AI-Video-Assistant/

A full-stack meeting intelligence tool that takes any YouTube URL or local video file and produces a complete analysis.

Pipeline:

  1. Audio extraction via yt-dlp + pydub (chunked into 10-min pieces)
  2. Transcription — English via OpenAI Whisper (local), Hinglish via Sarvam AI STT-Translate API
  3. Map-reduce summarization using Ollama phi3
  4. Extraction of action items, key decisions, and open questions
  5. RAG chat over the transcript using ChromaDB + HuggingFace embeddings + LCEL pipeline

Key Features:

  • Supports both YouTube URLs and local video/audio files
  • Bilingual: English + Hinglish (auto-translated to English via Sarvam)
  • Live pipeline status in sidebar with animated step indicators
  • Custom dark-themed Streamlit UI (JetBrains Mono + Syne fonts)
  • In-session RAG chat with full conversation history

Run:

streamlit run AI-Video-Assistant/app.py

2. Multi-Agent Research System

Multi-agent-research-system/

A 4-agent collaborative pipeline that researches any topic and produces a scored, structured report.

Agents:

Agent Role Tool
Search Agent Finds recent web information Tavily Search API
Reader Agent Scrapes top URLs for deep content BeautifulSoup scraper
Writer Chain Drafts structured research report Ollama phi3
Critic Chain Reviews and scores the report (X/10) Ollama phi3

Key Features:

  • Real-time pipeline status cards (Waiting → Running → Done)
  • Final report downloadable as .md
  • Critic feedback with score, strengths, and improvement areas
  • Custom dark UI with orange accent theme

Run:

# Streamlit UI
streamlit run Multi-agent-research-system/app.py

# CLI
python Multi-agent-research-system/pipeline.py

3. RAG Book Assistant

rag/

Upload any PDF and ask questions — answers grounded strictly in the document.

Architecture:

  • PDF ingestion via PyPDFLoader
  • Chunking with RecursiveCharacterTextSplitter (1000 chars, 200 overlap)
  • Embeddings via OllamaEmbeddings (nomic-embed-text)
  • Vector store: FAISS (persisted locally)
  • Retrieval: MMR (Maximal Marginal Relevance) — k=4, fetch_k=10
  • LLM: Ollama phi3 with strict context-only prompt

Also includes:

  • create_database.py — standalone script to pre-build FAISS index from a PDF
  • main.py — CLI query interface for the saved vector store
  • Experiments in retrievers/ — multi-query retriever, arXiv retriever, MMR tuning

Run:

streamlit run rag/app.py

4. Movie Recommender AI

movie-rec-ai/

A full-stack movie recommendation system with a FastAPI backend deployed on Render and a Streamlit frontend.

Backend (FastAPI):

  • TF-IDF cosine similarity on movie metadata (local dataset, ~45k movies)
  • TMDB API integration for posters, details, genre-based discovery
  • Endpoints: /home, /tmdb/search, /movie/id/{id}, /movie/search, /recommend/genre
  • Deployed at: https://movie-rec-466x.onrender.com

Frontend (Streamlit):

  • Keyword search with autocomplete dropdown
  • Poster grid with click-to-details navigation
  • Movie details page: poster, overview, genres, backdrop
  • Dual recommendations: TF-IDF similar movies + genre-based discovery

Run:

# Backend
uvicorn movie-rec-ai.main:app --reload

# Frontend
streamlit run movie-rec-ai/app.py

5. CineSage — Movie Info Extractor

cinesage/

Paste any movie description paragraph and get structured JSON output using Pydantic structured output with Ollama.

Output Schema:

class Movie(BaseModel):
    title: str
    release_year: Optional[int]
    genre: List[str]
    director: Optional[str]
    cast: List[str]
    rating: Optional[float]
    summary: str

Run:

streamlit run cinesage/UICore.py

6. AI Mode Chatbot

chatmodels/

A conversational chatbot with 5 distinct AI personality modes, built with full message history.

Modes: 😠 Angry · 😂 Funny · 😢 Sad · 😊 Happy · 🙄 Sarcastic

Key Features:

  • Persistent conversation memory using SystemMessage + HumanMessage + AIMessage
  • Mode switching resets chat context automatically
  • Message count stats in sidebar
  • Supports Ollama, OpenAI, Groq, Gemini (commented stubs included)

Run:

streamlit run chatmodels/UIChatbot.py

📚 Learning Modules

These modules document the learning journey through core LangChain concepts:

Module What's Covered
chatmodels/chat.py Basic LLM invocation across providers (Ollama, OpenAI, Groq, Gemini)
embeddingmodels/embedding.py Generating embeddings with Ollama (nomic-embed-text)
embeddingmodels/huggingfaceembedding.py HuggingFace sentence-transformers embeddings
tools/owntool.py Creating custom LangChain tools with @tool decorator
tools/toolcalling.py Manual tool binding + tool call loop with bind_tools
tools/Agents.py ReAct agent with weather + news tools + human approval middleware
tools/newssummarizer.py Tavily search → LCEL summarization chain
tools/sequencerunnable.py RunnableSequence (LCEL pipe operator)
tools/parallelrunnable.py RunnableParallel for concurrent chain execution
tools/runnablepassthrough.py RunnablePassthrough for context injection
rag/retrievers/mmr.py MMR retrieval tuning
rag/retrievers/multiquery.py Multi-query retriever for better recall
rag/retrievers/arixv.py ArXiv academic paper retriever

📋 Prerequisites

  • Python 3.11+
  • Ollama installed and running locally
  • Required Ollama models pulled: 
    ollama pull phi3
ollama pull nomic-embed-text
  • FFmpeg installed (for audio processing in AI Video Assistant)

🛠 Installation

# Clone the repository
git clone <repository-url>
cd "GenAI Learning"

# Install dependencies (using uv — recommended)
uv sync

# Or using pip
pip install -r requirements.txt

# Copy and fill environment variables
cp .env.example .env

🔑 Environment Variables

Create a .env file in the root directory:

# Ollama (local)
OLLAMA_BASE_URL=http://localhost:11434
OLLAMA_MODEL=phi3
OLLAMA_CHAT_MODEL=phi3:latest
OLLAMA_EMBEDDING_MODEL=nomic-embed-text:latest
OLLAMA_TEMPERATURE=0.3

# External APIs
TAVILY_API_KEY=<your-tavily-api-key>
TMDB_API_KEY=<your-tmdb-api-key>
OPENWEATHER_API_KEY=<your-openweather-api-key>
SARVAM_API_KEY=<your-sarvam-api-key>        # For Hinglish transcription

👤 Author

Sadik Mohammad

  • GitHub: [Your GitHub Profile]
  • Email: [Your Email]

Built with LangChain · Ollama · Streamlit · FastAPI · Whisper · FAISS · ChromaDB

About

This is My Machine Learning EMR Project

Resources

Readme

License

MIT license
Activity

Stars

0 stars

Watchers

0 watching

Forks

0 forks
Report repository

Releases

No releases published

Packages

 
 
 

Contributors

Languages