PodPilot - AI-Powered Podcast Recommendation System

Overview

PodPilot is an AI powered podcast recommendation system designed to help users discover new podcasts based on their preferences.

Link for video: https://www.linkedin.com/feed/update/urn:li:activity:7310988098996207617/

Why PodPilot?

Finding new podcasts that match your interests is often frustrating. Generic recommendations lead to repetitive content, making discovery time-consuming. PodPilot solves this problem by using AI based embeddings and search queries to enhance podcast discovery.

Tech Stack

1️⃣ Backend Framework

• FastAPI → High performance RESTful API framework
• Uvicorn → ASGI server to run FastAPI
• Postman → API testing

2️⃣ Cloud & Deployment

• Azure Container Apps → Cloud hosting
• Docker → Containerized FastAPI app
• GitHub Actions → CI/CD pipeline

3️⃣ APIs & External Integrations

• Taddy API → Fetches trending podcasts
• Google Custom Search API → Finds related podcasts
• OpenAI API → Improves text descriptions & generates search prompts
• Requests → Handles API calls

4️⃣ Databases & Data Handling

• SQLite → Stores podcasts & user preferences
• SQLAlchemy → ORM- Object Relational Mappe
• Fuzzy Matching (thefuzz) → Improves recommendations
• LangDetect → Detects podcast languages

5️⃣ Embeddings Creation

• Hugging Face Transformers Sentence-BERT (SBERT) → Generates embeddings & Converts text into numerical vectors

How It Works

1️⃣ Embeddings Creation Process

1. Fetch Podcast Data → Retrieves top podcast series from the Taddy API
2. Detect Language → Uses LangDetect to determine language
3. Optimize Descriptions → If Hebrew, translates to English via OpenAI GPT
4. Generate Embeddings → Converts podcast metadata into vectors using SBERT
5. Store Data → Saves podcasts & embeddings in SQLite

2️⃣ Recommendation Process

1. User Identification → User inputs their User ID
2. Preference Collection → User likes/dislikes podcasts to train the model
3. Search Query Generation → OpenAI GPT generates a Google Search API query
4. Find Related Podcasts → Scrapes Google Search API for similar podcasts
5. Personalized Recommendations → Embeddings & Fuzzy Matching return the top 3 podcasts

API Endpoints

Method	Endpoint	Description
POST	/suggest_podcasts	Suggests random podcasts from DB
POST	/process_feedback	Saves user preferences
POST	/generate_search_query	Creates a search query using OpenAI
POST	/search_related_podcasts	Finds podcasts via Google Search
POST	/recommend_podcasts	Returns the top 3 podcast recommendations