What I Built?

screenshot of web app deployed on render

I built Orbit for a close friend of mine in college.
He is funny, he is passionate, he is kind and he is really nice to be around. I adore him a lot. One of the biggest things on his bucket list for college was having a big friend group and getting over his stage fright. I wanted to give him a private, judgment free space where he could practice being the version of himself he wishes people could see.

Stage Mode: A private rehearsal stage where he can speak out loud without anyone watching. It analyzes speech pace, flags filler words, and gives encouraging, constructive feedback so he can walk into class with confidence.

Social Sandbox: Low stakes conversation simulations with AI personas. He can practice saying hello to a lab partner, joining a cafeteria conversation, or making friends in a crowded college fest.

Stress Radar: A quick daily check-in that spots when he starts pulling away. If his stress trends up for three consecutive days, Orbit gently nudges him with low effort ways to say hello to a friend, keeping him connected before the isolation spiral takes hold.

GitHub Repository

Mahi-Chandra / Orbit

Hacktoberfest DEV Weekend Challenge - Build for a friend

Orbit - Hacktoberfest 2026

AI-Powered Confidence & Connection Companion built for my friend
Submission for DEV Challenge: Build for a Friend (Hacktoberfest / Open-Source AI)

The Story: Why Orbit Exists

My friend has severe stage fright and his goal in college is to make lasting friends. He constantly looks for a circle, but whenever exams arrive or assignments pile up, he experiences extreme stress and pulls away into total isolation.
Orbit is designed to break this cycle. It acts as an empathetic, private space to practice speaking, simulate social interactions, and catch the warning signs of isolation before it takes over!

Core Pillars

1. Stage Mode

Real-time speech transcription via the browser-native Web Speech API (zero cloud voice tracking).
Automated linguistic diagnostics: filler word detection ("um", "like", "you know"), speaking speed (wpm), sentence complexity.
Compassionate AI evaluation scoring clarity, structure, and delivery confidence.

2. Social

…

View on GitHub

How I Built It?

Orbit is powered by open source models (Qwen 2.5 / Llama 3.3 open weights running via Groq for ultra-low latency inference)

Backend is built with FastAPI (Python 3.11) with a SQLite WAL-mode database for persistent tracking and metrics

Frontend has single page responsiveness using HTMX, Alpine.js, and Tailwind CSS

For voice and audio processing, I used Web Speech API for zero delay speech-to-text combined with Python based regex lexical tokenization for speech metrics and filler frequency analysis

Why Does Open Innovation Matter?

Social anxiety and stress tracking are deeply personal. Practicing an awkward speech rehearsal while admitting you feel panicky, or roleplaying reaching out to an estranged classmate are moments of vulnerability.

Closed, proprietary AI black boxes often treat user data as training fodder. This is why open innovation and open weight models matter! They provide you with-

Trust & Privacy: Open models allow users to see exactly what instructions are given and provide a foundation that can be run locally or within controlled environments where sensitive rehearsals never get commercialized.

Accessible Community Impact: Open source tools mean any student from anywhere in the world can clone this repository, run it for free, and adapt the social personas to their own regional culture or campus community.

DEV Community

I built a safe sandbox for my friend where he or anyone can practice small talk and public speaking without being judged