[11]competition

★ 2 hours · finished later

Byte the Vibe

Participant

Two hours to build a complete product, frontend and backend. Bhavishya and I built PlungePhoenix, an AI market-intelligence platform that listens to spoken market commentary, scores the panic in it and turns live financial news into trading insight. We didn't finish on the clock. I finished it afterwards. 🔥

When
May 2026
Where
Chitkara University, Punjab
Hosted by
Code 2 Chill Club · Chitkara University
Format
in person · 2 hours
The team at Byte the Vibe

[01]The squad

  • Govind Jindal← that's me
  • Bhavishya Grover

My role · Backend & AI pipeline

[02]The rounds

  1. Round 1

    The brief

    Build the whole thing, frontend and backend, in two hours. No mock-ups, no slides: a working product.

  2. Round 2

    The two hours

    I took the backend: FastAPI, speech-to-text, the panic scorer and the RAG pipeline, with the audio input fixed at 17:23. The frontend landed in plain HTML, CSS and JS at 17:53, when I'd planned for React, and the time went into stitching the two together. The clock won.

    Not finished on time ⏱️

  3. Round 3

    After the bell

    In July I came back and finished it properly: a React 19 frontend, API retries, structured errors, a Gemini fallback for the LLM and a full SaaS landing page, deployed on Vercel.

    Shipped 🚀

[03]What we built

PlungePhoenix

PlungePhoenix home page

Running live, right here. Click around.

  • Audio sentiment analysis

    Record or upload market commentary. AssemblyAI transcribes it, with exponential-backoff retries, and the transcript drives everything after.

  • A panic score

    A weighted keyword engine ("crash" counts for 1.0, "volatile" for 0.5) scores market stress from 0 to 1 and classifies it from low to extreme, streamed live over a WebSocket.

  • RAG over live news

    Financial news from RSS feeds is cleaned, embedded and stored in ChromaDB. The transcript pulls back the ten most relevant articles as context.

  • AI recommendations

    Transcript, articles and panic score go to an LLM that returns structured buy, sell or hold signals with confidence and the sectors affected.

  • An LLM that doesn't go down

    OpenRouter first, Gemini as the fallback, with retries and automatic re-prompting when the model returns malformed JSON.

FastAPIPythonReact 19ViteAssemblyAIChromaDBOpenRouterGeminiWebSockets

Live siteGitHub

[04]Plot twist

The stack decision

Then 😱

Split the work and start typing: I'd do the backend, my partner the frontend. I wanted React; the frontend came back in HTML, CSS and JS, and two hours don't leave room to rebuild or to glue mismatched pieces.

Now 🧠

Agree on the stack before the clock starts, in the first five minutes, out loud. The mistake cost us the finish, and taught me more than a finish would have.

[05]The two hours, in commits

Commits
6
Span
62 min
Avg. message
25.5 chars
Shortest
“Added Run File”

Times in IST, from the repo. The rest of the history, from 25 July, is the version you can open today.

[06]From the floor

The analysis dashboard
The analysis dashboard

[07]The certificate

Certificate for Byte the VibeDelivered ✓

Certificate found. 📬

For participating in Byte The Vibe, organised by the Code 2 Chill Club on 18 May 2026. Signed by Dr. Manav Bansal, Chief Happiness Officer, Office of Student Life. 📜

“Once again the clock won, and the lesson was clear: the code wasn't the bottleneck, the coordination was. Pick the stack together, then build. And finish the thing anyway. 🔥”

Next hackathon

Finvasia Innovation Hackathon 2026

← All hackathons