ReelPilot AI: The Ghost in Your Audience 👻
An AI-powered Instagram audience intelligence tool that turns the voices hidden in comments into ideas for the next Reel.
What I Built
Every Instagram comment is a small signal. A question, a suggestion, a complaint, or an idea someone wants to see next.
But when a creator has hundreds or thousands of comments, finding those signals manually becomes difficult.
I built ReelPilot AI to solve that problem.
ReelPilot analyses Instagram posts and audience comments to discover what people are actually interested in. Instead of suggesting generic trending topics, it looks for patterns in the creator's own audience.
It can:
- Connect to an Instagram account and import comments.
- Analyse comments and identify recurring questions and content requests.
- Group similar comments into topic clusters.
- Rank audience signals using frequency, relevance, and other signals.
- Generate Reel ideas based on actual audience demand.
- Turn an audience signal into an editable Reel content pack.
For example, in my demo, ReelPilot analyses 603 imported comments and surfaces audience questions, topic clusters, and ready-to-make Reel ideas.
How does this connect to Ghost in the Machine?
I interpreted the theme a little differently.
The ghost isn't a fictional character. It's the collective voice of an audience hiding inside hundreds of comments.
Every comment is a fragment of what someone wants to know, learn, or watch. ReelPilot brings those voices together and helps creators hear what their audience is trying to tell them.
The idea is simple: don't guess what your audience wants. Listen to it.
Demo
The application is not publicly deployed yet. I'm sharing a recorded walkthrough demonstrating the working application, Instagram connection, comment analysis, audience signals, and Reel content generation.
Video demo: [https://drive.google.com/drive/folders/126dj8djPnJc3tKfffzBOdPUbsXI2mulr?usp=sharing]
The screenshots attached to this post show the ReelPilot dashboard and the content-generation workflow.
Code
GitHub Repository:
https://github.com/kavyanshpandey/ReelPilotAI
The project is built as a web application with a frontend and backend.
How I Built It
The most important part of ReelPilot is that its AI intelligence is powered by an open-weight model rather than depending entirely on a closed AI API.
AI: Gemma + Ollama
I use Google Gemma, served locally through Ollama on my MacBook.
The model powers the AI-driven content understanding and generation workflow. It helps transform audience signals into useful, structured Reel ideas and content suggestions.
Running Gemma locally gives me control over the inference environment and lets me experiment with an open-weight model without making a hosted, proprietary AI API a mandatory part of the generation workflow.
Database: MongoDB Atlas
I use MongoDB Atlas as the database layer for application data.
It provides the persistence layer needed to organise creator information, imported audience data, analysis results, and generated content.
Application
The application brings together:
- Instagram integration for retrieving audience comments.
- A backend processing pipeline for cleaning, analysing, and grouping comments.
- Audience intelligence to surface recurring questions and topic opportunities.
- Gemma-powered content generation.
- A dashboard where creators can review insights and work with generated ideas.
The goal was not to build another AI text generator. It was to build a workflow where the AI has real audience context before it suggests what a creator should make next.
Why Does Open Innovation Matter?
For me, open innovation means having the freedom to choose how AI works inside a product.
With Gemma running locally through Ollama, I can experiment with the model on my own machine instead of making every generation dependent on a closed, external AI service.
This matters for three reasons:
1. More control over data and inference
Audience comments can contain personal opinions and information. Running the model locally means the AI inference step does not need to send that text to a third-party model API. MongoDB Atlas remains a separate cloud-hosted data layer, so this is not a claim that the entire application is offline.
2. Freedom to experiment
Open-weight models make it possible to test different model configurations, prompts, and workflows. I can explore how the model handles English, Hindi, and Hinglish audience conversations and improve the results for creators.
3. Less dependence on proprietary AI services
I don't want the core intelligence of ReelPilot to be permanently tied to one closed AI provider. Open models give developers more room to experiment, adapt, and build products on their own terms.
Open innovation made it possible for me to build and experiment with an AI-powered creator tool using a model that I can run and control locally.
The Spirit of It
A creator can spend hours making a Reel, but the next great idea might already be sitting in the comments of their previous post.
Buried among emojis, jokes, questions, and random conversations are people asking for something more.
ReelPilot is my attempt to bring those forgotten voices back to the surface.
The ghost in this machine is the audience itself. 👻
Prize Categories
- Best Use of Gemma (Featured category)
- Best Use of MongoDB Atlas (Partner category)
About the Builder
Built individually by Kavyansh Pandey.
Top comments (0)