Skip to content
BranchesTags

Repository files navigation

GenAI Learning

Hands-on Generative AI projects spanning LangChain, LangGraph, RAG, agents, MCP, and computer vision.

A learning-focused monorepo of Python experiments and demo apps—from basic LLM calls and embeddings to multi-agent research, travel planning with MCP, video meeting intelligence, and content-based movie recommendations.

Python LangChain LangGraph Streamlit FastAPI License: MIT GitHub


🚀 Live Demos

AI Video Assistant   Multi-Agent Research   TripMate AI

RAG Book Assistant   Movie Recommender   CineSage

App Live URL
🎬 AI Video Assistant genai-ai-video-assistant.streamlit.app
🔬 Multi-Agent Research genai-multiagent-research-ai.streamlit.app
✈️ TripMate AI tripmates-ai.netlify.app
📚 RAG Book Assistant genai-rag-ai.streamlit.app
🎞️ Movie Recommender genai-movie-rec-ai.streamlit.app
🍿 CineSage Extractor genai-movie-extractorai.streamlit.app

Table of Contents

Overview

GenAI Learning is a personal learning repository that explores modern generative AI patterns in Python. Each folder is a self-contained experiment or demo app—not a single production product.

It addresses common GenAI learning goals:

Goal How this repo explores it
Talk to local LLMs Ollama (phi3) via LangChain in chat, RAG, and agent demos
Ground answers in documents PDF → chunk → embed → retrieve (FAISS / Chroma)
Orchestrate multi-step agents LangGraph graphs, tool-calling agents, critic loops
Connect external tools Tavily, OpenWeatherMap, TMDB, AviationStack, Sarvam STT, MCP servers
Ship simple UIs / APIs Streamlit frontends and FastAPI backends

This is an educational codebase. Features, dependencies, and run instructions vary by subproject. Some apps expect local services (Ollama, FFmpeg, PostgreSQL) and API keys that are not bundled with the repo.

Key Features

  • AI Video Assistant — YouTube / local video → audio chunks → Whisper or Sarvam transcription → map-reduce summary → action items / decisions / questions → Chroma RAG chat
  • Multi-Agent Research System — Search (Tavily) → scrape → write report → critic score (Ollama phi3), with Streamlit UI and CLI
  • TripMate AI — LangGraph travel planner with MCP-backed flight, hotel, and weather tools; FastAPI + HTML UI; PostgreSQL checkpoints
  • College Assistant (Agentic AI/) — LangGraph router over academic / fee PDFs with Groq + FAISS + HuggingFace embeddings
  • RAG Book Assistant — Upload a PDF, build a FAISS index, ask questions with MMR retrieval and Ollama
  • Movie Recommender AI — FastAPI (TF-IDF over ~45k movies + TMDB) and Streamlit UI (default API host on Render)
  • CineSage — Structured movie metadata extraction with Pydantic + Ollama
  • AI Mode Chatbot — Streamlit chatbot with five personality modes and message history
  • Learning modules — LCEL runnables, tool calling, embeddings, retrievers, LangGraph patterns, MediaPipe hand/pose tracking

Architecture / Workflow

Repository map

flowchart TB
  subgraph apps [Demo Applications]
    AVA[AI-Video-Assistant]
    MARS[Multi-agent-research-system]
    TM[TripMate-AI-Using-MCP]
    RAG[rag]
    MR[movie-rec-ai]
    CS[cinesage]
    CB[chatmodels]
    AA[Agentic AI]
  end

  subgraph learn [Learning Modules]
    TOOLS[tools]
    EMB[embeddingmodels]
    CV[Computer_Vision]
  end

  LLM[(Ollama / Groq / OpenAI)]
  VS[(FAISS / Chroma)]
  APIs[(Tavily · TMDB · Weather · AviationStack · Sarvam)]

  AVA --> LLM
  AVA --> VS
  AVA --> APIs
  MARS --> LLM
  MARS --> APIs
  TM --> LLM
  TM --> APIs
  RAG --> LLM
  RAG --> VS
  AA --> LLM
  AA --> VS
  MR --> APIs
  CS --> LLM
  CB --> LLM
  TOOLS --> LLM
  TOOLS --> APIs
Loading

AI Video Assistant pipeline

flowchart LR
  A[YouTube URL / local file] --> B[yt-dlp + pydub]
  B --> C[10-min WAV chunks]
  C --> D{Language}
  D -->|english| E[OpenAI Whisper]
  D -->|hinglish| F[Sarvam STT-Translate]
  E --> G[Transcript]
  F --> G
  G --> H[Summary · Title · Extractions]
  G --> I[Chroma + HF embeddings]
  I --> J[LCEL RAG chat]
Loading

Multi-agent research pipeline

flowchart LR
  T[Topic] --> S[Search Agent + Tavily]
  S --> R[Reader Agent + scrape_url]
  R --> W[Writer Chain]
  W --> C[Critic Chain]
  C --> O[Report + Score X/10]
Loading

TripMate LangGraph flow

flowchart LR
  U[User message] --> F[flight_agent]
  F --> H[hotel_agent]
  H --> W[weather_agent]
  W --> I[itinerary_agent]
  I --> X[final_agent]
  X --> Y[JSON response]
Loading

Tech Stack

Category Technologies used in this repo
Language Python (pyproject.toml: >=3.13; .python-version: 3.13)
LLM frameworks LangChain, LangGraph, LangChain Community / HuggingFace / text-splitters
Local LLMs Ollama — typically phi3, embeddings nomic-embed-text
Hosted LLMs Groq (llama-3.3-70b-versatile), OpenAI (selected LangGraph demos)
Vector stores FAISS (faiss-cpu), Chroma (langchain-chroma)
Embeddings Ollama Embeddings, HuggingFace (all-MiniLM-L6-v2, mixedbread-ai/mxbai-embed-xsmall-v1)
Tools / APIs Tavily, OpenWeatherMap, TMDB, AviationStack, Sarvam AI STT
MCP langchain-mcp-adapters, mcp, Tavily MCP HTTP, aviationstack-mcp via uvx, custom weather MCP server
Audio / video OpenAI Whisper, yt-dlp, pydub, FFmpeg
Web Streamlit, FastAPI, Uvicorn, Jinja2 templates
ML / CV scikit-learn (TF-IDF), OpenCV, MediaPipe
Persistence FAISS local indexes, Chroma dirs, PostgreSQL (TripMate checkpoints via langgraph-checkpoint-postgres)
Package managers uv (pyproject.toml / uv.lock) and pip (requirements.txt; TripMate has its own)

Project Structure

GenAI Learning/
├── AI-Video-Assistant/           # Meeting / video intelligence (Streamlit + CLI)
│   ├── app.py
│   ├── main.py
│   ├── core/                     # transcriber, summarizer, extractor, RAG, vector store
│   └── utils/audio_processor.py
├── Multi-agent-research-system/  # Search → scrape → write → critique
│   ├── app.py                    # Streamlit UI
│   ├── pipeline.py               # CLI runner
│   ├── agents.py
│   └── tools.py
├── TripMate-AI-Using-MCP/        # Multi-agent travel planner (FastAPI + MCP)
│   ├── app.py
│   ├── backend.py
│   ├── mcp_client.py
│   ├── custom_weather_mcp_server.py
│   ├── templates/
│   ├── tools/
│   └── requirements.txt
├── Agentic AI/                   # LangGraph learning + College Assistant UI
│   ├── app.py                    # Streamlit college assistant
│   ├── sequential_base.py
│   ├── conditional_RAG.py
│   ├── parallel_reducers.py
│   ├── humanintheloop.py
│   ├── iterative_tools.py
│   └── states.py
├── rag/                          # PDF RAG (Streamlit + CLI helpers)
│   ├── app.py
│   ├── create_database.py
│   ├── main.py
│   ├── document loaders/
│   ├── retrievers/
│   └── vector store/
├── movie-rec-ai/                 # TF-IDF + TMDB recommender
│   ├── main.py                   # FastAPI
│   ├── app.py                    # Streamlit
│   ├── movies.ipynb
│   └── *.pkl                     # Precomputed TF-IDF artifacts
├── cinesage/                     # Structured movie extraction
│   ├── core.py                   # CLI (ollama chat)
│   └── UICore.py                 # Streamlit
├── chatmodels/                   # Chat experiments + mode chatbot
├── tools/                        # Tools, agents, LCEL runnables
├── embeddingmodels/              # Embedding experiments
├── Computer_Vision/              # MediaPipe hand / pose demos
├── main.py                       # Prints langchain.__version__
├── pyproject.toml
├── requirements.txt
├── uv.lock
├── LICENSE
└── README.md

Prerequisites

  • Python 3.13+ recommended for the root project (pyproject.toml / .python-version)
  • Ollama running locally for Ollama-based demos, with models pulled as needed:
ollama pull phi3
ollama pull nomic-embed-text
  • FFmpeg on PATH for AI Video Assistant audio extraction
  • API keys for the apps you run (see Configuration)
  • PostgreSQL (and DATABASE_URL) for TripMate conversation checkpoints
  • uvx available if you use TripMate’s AviationStack MCP (uvx aviationstack-mcp)
  • Webcam / OpenCV for Computer Vision scripts

Installation

git clone https://github.com/SadiqCodex/GenAI-Learning-Project.git
cd GenAI-Learning-Project

Option A — uv (root project)

uv sync

Option B — pip (root requirements.txt)

python -m venv .venv

# Windows
.venv\Scripts\activate

# macOS / Linux
source .venv/bin/activate

pip install -r requirements.txt

TripMate-specific dependencies

pip install -r TripMate-AI-Using-MCP/requirements.txt

Additional packages used by some apps

Root pyproject.toml / requirements.txt do not list every import used across subprojects. Depending on what you run, you may also need packages such as:

langchain-ollama, langchain-chroma, langchain-groq, openai-whisper, yt-dlp, pydub, httpx, tavily-python, beautifulsoup4

Install only what the target script imports.

Create a root .env (there is no .env.example in the repo):

# Create manually, then edit values
# Windows PowerShell:  New-Item .env
# macOS / Linux:       touch .env

Configuration / Environment Variables

Place a .env in the repository root. Most modules call load_dotenv() and resolve keys from there (some also check a local .env beside the script).

Variable Used by Purpose
OLLAMA_BASE_URL Many Ollama demos Default http://localhost:11434
OLLAMA_MODEL Research system, video cores Default phi3
OLLAMA_CHAT_MODEL RAG, tools Default phi3:latest
OLLAMA_EMBEDDING_MODEL RAG Default nomic-embed-text:latest
OLLAMA_TEMPERATURE Video / research Float temperature
WHISPER_MODEL AI Video Assistant Whisper size (default small)
SARVAM_API_KEY AI Video Assistant (Hinglish) Sarvam STT-Translate
SARVAM_STT_MODEL AI Video Assistant Default saaras:v2.5
TAVILY_API_KEY Research, tools, TripMate Web search
TMDB_API_KEY movie-rec-ai Movie metadata / posters
OPENWEATHER_API_KEY tools/Agents.py, TripMate weather MCP Weather
GROQ_API_KEY TripMate, Agentic AI demos Groq chat models
OPENAI_API_KEY Selected LangGraph scripts e.g. human-in-the-loop writer
DATABASE_URL TripMate PostgreSQL for LangGraph checkpoints
AVIATIONSTACK_API_KEY TripMate Flight search / MCP
DEFAULT_ORIGIN_IATA TripMate flight tool Default origin (code default DAC)
HUGGINGFACEHUB_API_TOKEN Optional HF usage Present in local env patterns
GEMINI_API_KEY / OPENROUTER_API_KEY / MISTRAL_API_KEY Optional / experiment May appear in .env; not required for every app

Example skeleton (replace placeholders; do not commit real secrets):

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

TAVILY_API_KEY=
TMDB_API_KEY=
OPENWEATHER_API_KEY=
GROQ_API_KEY=
OPENAI_API_KEY=
SARVAM_API_KEY=

DATABASE_URL=postgresql://user:password@localhost:5432/travel_db
AVIATIONSTACK_API_KEY=
DEFAULT_ORIGIN_IATA=DAC

.env is listed in .gitignore and should stay out of version control.

Usage

Run commands from the repository root unless noted. Activate your virtual environment first.

AI Video Assistant

streamlit run AI-Video-Assistant/app.py

# CLI
python AI-Video-Assistant/main.py

Provide a YouTube URL or local media path; choose english (Whisper) or hinglish (Sarvam).

Multi-Agent Research System

streamlit run Multi-agent-research-system/app.py
python Multi-agent-research-system/pipeline.py

TripMate AI

python TripMate-AI-Using-MCP/app.py

Open http://127.0.0.1:8000/. Update the weather MCP command / args paths in TripMate-AI-Using-MCP/mcp_client.py to your machine before relying on weather tools.

College Assistant (Agentic AI)

streamlit run "Agentic AI/app.py"

Expects academics_handbook.pdf and fee_structure.pdf beside the app (both are present in Agentic AI/).

RAG Book Assistant

streamlit run rag/app.py

# Pre-build FAISS from bundled sample PDF, then query via CLI
python rag/create_database.py
python rag/main.py

Movie Recommender

# API (requires TMDB_API_KEY and pickle artifacts in movie-rec-ai/)
uvicorn movie-rec-ai.main:app --reload

# UI (defaults to hosted API base URL in app.py)
streamlit run movie-rec-ai/app.py

CineSage

streamlit run cinesage/UICore.py
python cinesage/core.py

AI Mode Chatbot

streamlit run chatmodels/UIChatbot.py

Modes: Angry · Funny · Sad · Happy · Sarcastic.

Learning scripts (examples)

python chatmodels/chat.py
python embeddingmodels/embedding.py
python tools/toolcalling.py
python tools/Agents.py
python "Agentic AI/sequential_base.py"
python Computer_Vision/HandTracking.py

API Reference

Movie Recommender (movie-rec-ai/main.py)

Method Path Description
GET /health Health check
GET /home TMDB feed (category: trending, popular, top_rated, upcoming, now_playing)
GET /tmdb/search TMDB keyword search
GET /movie/id/{tmdb_id} Movie details
GET /movie/search Search bundle: details + TF-IDF + genre recommendations
GET /recommend/tfidf TF-IDF similar titles
GET /recommend/genre Genre-based TMDB discovery

Streamlit frontend currently sets API_BASE to https://movie-rec-466x.onrender.com (with a local URL as a non-effective fallback expression in code). Point it at http://127.0.0.1:8000 when testing the API locally.

TripMate (TripMate-AI-Using-MCP/app.py)

Method Path Description
GET / Web UI
GET /health Health check
POST /api/travel Travel plan request

Request body:

{
  "message": "Plan a 3-day trip to Tokyo with a budget of $1200",
  "thread_id": null
}

Example:

curl -X POST http://127.0.0.1:8000/api/travel \
  -H "Content-Type: application/json" \
  -d "{\"message\":\"Plan a 3-day trip to Tokyo with a budget of $1200\"}"

Examples

CineSage structured schema (cinesage/UICore.py / core.py):

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

RAG retrieval settings (rag/app.py): chunk size 1000, overlap 200, MMR with k=4, fetch_k=10, lambda_mult=0.5.

Movie dataset artifact: movie-rec-ai/df.pkl holds 45,447 rows with columns including title, overview, genres, tagline, vote_average, popularity, tags.

Testing

There is no automated pytest suite in this repository. Manual / exploratory scripts include:

File Role
AI-Video-Assistant/test.py Manual pipeline smoke run
TripMate-AI-Using-MCP/test.py Manual TripMate checks
TripMate-AI-Using-MCP/mcp_client_test.py MCP client experiment
rag/document loaders/test.py Document-loader experiment

pytest appears in requirements.txt but no test_*.py unit tests are defined for CI.

Deployment

Hosted apps

Project Platform Live demo
AI Video Assistant Streamlit Cloud Open demo
Multi-Agent Research Streamlit Cloud Open demo
RAG Book Assistant Streamlit Cloud Open demo
Movie Recommender Streamlit Cloud Open demo
CineSage Extractor Streamlit Cloud Open demo
TripMate AI Netlify Open demo

Other deployment notes

Component Notes present in repo
Movie API movie-rec-ai/runtime.txt specifies python-3.11.9 (Render-style). Streamlit UI references https://movie-rec-466x.onrender.com.
TripMate DATABASE_URL error text mentions a Render PostgreSQL external URL; local run uses Uvicorn on 127.0.0.1:8000. Frontend is also hosted at tripmates-ai.netlify.app.

No Dockerfiles or CI workflows are included in this repository.

Troubleshooting

Issue What to check
Connection refused to Ollama Ollama running; OLLAMA_BASE_URL; models pulled (phi3, nomic-embed-text)
TMDB_API_KEY missing Set TMDB_API_KEY in root .env before importing movie-rec-ai.main
SARVAM_API_KEY is not set Required for hinglish transcription
DATABASE_URL is missing / Groq errors Required for TripMate (backend.py)
TripMate weather MCP fails mcp_client.py currently hardcodes a machine-specific Python path and script path—update both to your local custom_weather_mcp_server.py
Audio / YouTube failures Install FFmpeg; confirm yt-dlp / pydub imports
FAISS / Chroma lock on Windows RAG app retries directory removal and may fall back to a timestamped folder
ModuleNotFoundError Install the missing package for that subproject (root lockfile does not cover every demo)
Movie UI cannot reach API Confirm API is up, or change API_BASE in movie-rec-ai/app.py to your local Uvicorn URL

Future Improvements

Ideas aligned with the current codebase gaps:

  • Add a root .env.example documenting all keys without secrets
  • Align pyproject.toml / uv.lock / requirements.txt with packages actually imported by each app
  • Add a small pytest smoke suite for FastAPI /health routes and pure helpers
  • Make TripMate MCP weather paths configurable via environment variables
  • Add a root static/ mount or serve TripMate CSS/JS consistently with FastAPI
  • Optional Docker Compose for Ollama + PostgreSQL + selected APIs

Contributing

Contributions that improve clarity, dependency hygiene, or demos are welcome.

  1. Fork SadiqCodex/GenAI-Learning-Project
  2. Create a feature branch
  3. Keep changes scoped; do not commit .env or API keys
  4. Open a pull request with a short description of what you ran and how you verified it

License

This project is licensed under the MIT License — see LICENSE.

Copyright (c) 2025 Sadik Mohammad.

Contact

Sadik Mohammad (SadiqCodex)

🌐 Try the Live Apps

Click a badge to open the deployed demo

AI Video Assistant   Multi-Agent Research   TripMate AI

RAG Book Assistant   Movie Recommender   CineSage

# Project Live demo
1 AI Video Assistant https://genai-ai-video-assistant.streamlit.app/
2 Multi-Agent Research System https://genai-multiagent-research-ai.streamlit.app/
3 TripMate AI http://tripmates-ai.netlify.app/
4 RAG Book Assistant https://genai-rag-ai.streamlit.app/
5 Movie Recommender AI https://genai-movie-rec-ai.streamlit.app/
6 CineSage Movie Extractor https://genai-movie-extractorai.streamlit.app/

Built as a hands-on GenAI learning lab — LangChain · LangGraph · Ollama · Streamlit · FastAPI · Whisper · FAISS · Chroma · MCP

About

A hands-on learning repository for building and understanding agentic AI systems, LLM applications, AI agents, workflows, RAG, tool calling, and modern AI engineering concepts.

genai-multiagent-research-ai.streamlit.app/

Topics

agentic-aiagentic-workflowsai-agentsai-engineeringai-workflowsartificial-intelligenceautonomous-agentsdeep-learninggenerative-ailangchainlanggraphlarge-language-modelsllmllm-applicationsmachine-learningprompt-engineeringpythonragretrieval-augmented-generationtool-calling

Resources

Readme
MIT license
Activity

Stars

0 stars

Watchers

0 watching

Forks

0 forks
Report repository

Releases

Packages

Contributors

Languages