PocketWisdom AI is a RAG-powered educational chatbot that helps parents explain school topics to their children in fun, kid-friendly language. It retrieves knowledge from YouTube transcripts, finds relevant diagrams, and presents simplified educational responses — all driven by GPT-4o-mini.
This project was developed as part of the Ironhack final capstone and uses LangChain, LangSmith, and vector search to build a truly helpful educational assistant.
Agent-Powered Tool Usage:
🧾 Document Search: Retrieves relevant YouTube transcript chunks
🔢 Calculator: Performs step-by-step numeric reasoning
🖼️ Image Search: Auto-fetches diagrams or visuals related to the topic (no fallbacks used)
GPT-4o-mini: Fast, accurate, and cost-effective reasoning and explanation
LangSmith Evaluation: Validates output accuracy, hallucination, and relevance
Streamlit App: Friendly local interface for natural language chat
ParentWise/
├── faiss_youtube_index/ # Saved FAISS vector store for fast retrieval
│ ├── index.faiss # FAISS index file
│ └── index.pkl # Metadata and vector data
│
├── app/ # Main app directory
│ ├── app.py # Streamlit app interface
│ ├── Logo
│ └── faiss_youtube_index/ # Local copy of the FAISS index for the app
│
├── youtube_transcripts/ # Educational text data used to build the index
│ ├── Art.txt
│ ├── Biology.txt
│ ├── Chemistry.txt
│ ├── Computer_Science.txt
│ ├── Geography.txt
│ ├── History.txt
│ └── Physics.txt
│
├── Evaluation.ipynb # Notebook for LangSmith-based evaluation
├── VectotStore+Agent&tools.ipynb # Tool definitions, retriever and agent setup
├── YouTube_decipher.ipynb # Script for extracting YouTube content
├── requirements.txt
└── .env
LangChain + Agents
OpenAI GPT-4o-mini
FAISS – Vector DB for transcript retrieval
DuckDuckGo Search – Safe and simple image fetch
Streamlit – Local chat app interface
LangSmith – Evaluation with LLM-based graders
Python 3.10+
Install all required Python packages with: (pip install -r requirements.txt)
This application runs locally using Streamlit. To launch the app: (streamlit run app.py)
This project is licensed under the MIT License.
rag-answer-vs-reference - Compares the AI answer to a reference answer for factual correctness
Huge thanks to my teachers for their support, guidance, and encouragement throughout this journey. Big appreciation to my Ironhack classmates for their feedback and collaboration. Acknowledgment is extended to Ironhack for the structure, tools, and opportunity.