AI · Data Case Study

Community
Sentiment Dashboard

A sentiment analysis tool built for Telegram communities. No engineers, just a problem worth solving.

Skip to live tool
Context
The situation
Organization
Stacks Asia Foundation
Deliverable
Free alternative to paid community analytics
Stack
Python, HuggingFace, Google Sheets, HTML/JS
Deployment
Hosted on Render, runs locally
The problem
Flying blind on community health

Stacks' Telegram community was active, multilingual, and impossible to monitor manually. Complaints repeated, sentiment shifted, and there was no free tool that could surface any of it.
So I built one.

Live tool — illustrative demo with synthetic data Open in full screen
Launch live tool
Hosted on Render.
Takes ~30 seconds to wake up on first load.
What I built
End-to-end, no engineers required
01 Designed the full system architecture independently with Cursor, deciding on a local Python pipeline for privacy with a hosted web viewer for easy sharing.
02 Built a multilingual NLP pipeline using HuggingFace sentiment models (a Korean-specific model and a standard English model) processing Telegram JSON exports locally.
03 Designed and built the full dashboard in HTML/JS with four views covering top complaints, sentiment trends over time, and top community voices.
04 Built and deployed a public illustrative demo using 90,000 synthetic messages, so the tool can be shared without exposing real user data.
Python HuggingFace Transformers Google Sheets API HTML / CSS / JavaScript Render Telegram JSON export
Outcomes
What it delivered for my team
Early crisis signals Caught sentiment spikes after protocol events before they became community issues.
Event impact visibility Tracked how announcements and launches moved community sentiment.
Repeat complaints surfaced Identified recurring friction points that would have gone unnoticed.
Proactive strategy Acted on data rather than reacting to complaints after the fact.
Bilingual coverage Monitored Korean and English channels with equal accuracy.
Zero cost Replaced the need for a paid sentiment tool with a free alternative.
How to use it
Try it for yourself!

Run the tool with your own community data. It runs locally. Your messages stay on your machine.

1
Download from GitHub
Clone the repo or download as a ZIP.
2
Install Python dependencies
Requires Python 3.9+.
3
Export your Telegram chat
Use Telegram Desktop to export as JSON.
4
Run sentiment analysis
First run downloads the AI model (~250MB).
5
Open the dashboard
Start the local server or connect via Google Sheets.
6
Explore your data
Browse Dashboard, Trends, and Complaints.
Full setup guide on GitHub
Previous project
Bitcoin's Missing Layer
View case study