💬 WhatPeopleSay

Live Website: whatpeoplesay.onrender.com

---

What is this project?

When you search for something on Google, you mostly get news articles, blogs, and advertisements.

WhatPeopleSay is a different kind of search engine. It searches through thousands of real, raw comments and discussions posted by everyday people online. It then uses Artificial Intelligence (AI) to read all those comments, summarize them, and show you exactly what people's opinions, experiences, and advice are on that topic.

Basically, it reads through hundreds of posts for you and gives you a neat, honest report of "what people are saying" in seconds.

---

Technical Details

This project is built using a clean Python backend and a vanilla web frontend. Here is how the technical components work together:

• Database (SQLite): A lightweight SQL database stores raw discussion data and comments. We use SQLite's FTS5 extension to handle fast keyword matching.
• Hybrid Search (Reciprocal Rank Fusion): To find the most relevant posts, the search engine combines two methods:
  • Full-Text Search: Finds exact matches of the keywords you typed.
  • Vector Semantic Search: Uses a Machine Learning model (sentence-transformers/all-MiniLM-L6-v2) to convert comments and your search query into coordinates (embeddings) to find matches based on meaning, even if different words are used.
  • The results are merged and ranked using Reciprocal Rank Fusion (RRF) with a weight distribution (HYBRID_ALPHA = 0.65).
• AI Summarization (Groq API): The top matching posts are compiled and sent to Groq's Llama-3 model (llama-3.3-70b-versatile). The system guides the AI to:
  • Base the summary only on the provided text (no hallucinated details).
  • Mask all usernames with neutral labels (like "a commenter").
  • Output a formatted layout with a one-line answer, a deep summary, bullet takeaways, and sources.
• Web Frontend (FastAPI Templates): Served as a single, self-contained HTML page using FastAPI's Jinja2 template engine. It handles:
  • Tabbed mobile views (switching between Insights and Context).
  • Clickable references that scroll to and highlight source cards.
  • Collapsible sources toggle blocks.
  • Custom pure CSS Apple-style loader spinner animations.

---

Key Features

• Honest Summaries: Uses AI to write direct, unbiased summaries of public discussions.
• Original Sources: Every summary includes a toggle button to see the links to the original posts.
• Clickable References: Click on names inside the summary to jump straight to their posts.
• Apple Loading Spinners: Beautiful iOS-style loading animations.
• Perfect on Mobile: Automatically fits on phone screens with a dual-tab layout.

---

How to Run It Locally

1. Install dependencies:

   pip install -r requirements.txt

2. Start the server:

   python -m uvicorn backend.app:app --host 127.0.0.1 --port 8000

3. Open in browser: Go to: http://127.0.0.1:8000

---

Deploying to Render

This project contains a render.yaml file. You can deploy it instantly:

1. Upload your code to GitHub.
2. Select Blueprint on Render (dashboard.render.com).
3. Connect your repository, add your Groq API keys, and click deploy!

---

Credits

Created by Afroz Alam

• Email: httppirate.protonmail.com