DEV Community

Gourav Chhatwani
Gourav Chhatwani

Posted on AI-assisted

TinySteps AI: Building an AI-Powered Task Roadmap Generator with Local Google Gemma, Streamlit, and Full CI/CD

#devchallenge #hf26challenge

⚡ TinySteps AI: Building an AI-Powered Task Roadmap Generator with Local Google Gemma, Streamlit, and Full CI/CD

Hey everyone! 👋

For Week 3 of Hacktoberfest, I decided to tackle a problem I deal with almost every single day: jumping into a big, intimidating project—like building an ML pipeline from scratch or setting up microservices—and feeling completely paralyzed by where to start.

To solve that, I built TinySteps AI. It’s a tool that takes any complex software or ML engineering objective and breaks it down into structured, realistic, energy-aware phases you can actually execute without burning out.

Check out the full walkthrough video below to see it generating roadmaps in real-time and triggering our automated deployment pipeline:

🎯 What Does TinySteps AI Do?

Instead of just spitting out a generic checklist, TinySteps AI adapts to your real-world setup:

  1. Energy-Aware Planning: Select your current energy level (Low, Medium, or High) along with your available time budget (10 to 240 mins). The engine dynamically adjusts phase durations and depth so you don't overcommit when you're tired.
  2. Interactive Streamlit Frontend: Clean, fast dashboard built with Streamlit for tweaking inputs and viewing instant execution blueprints.
  3. Dual-Backend AI Engine:
    • Local Inference (Google Gemma): Runs gemma3:4b locally via Ollama. It keeps your code ideas completely private, works offline, and costs $0.
    • Cloud Fallback (OpenRouter): Integrates free cloud endpoints so the app stays functional anywhere.

🏗️ Under the Hood: Full CI/CD & Production Setup

I wanted this to be a production-ready application rather than just a local script, so I packed in several devops and pipeline features:

  • Automated Testing: Unit test suite using pytest verifying engine fallback logic, step sizing, and key mapping.
  • GitHub Actions CI/CD: Runs automated tests on every push to main, builds a Docker container, and publishes the image directly to the GitHub Container Registry (GHCR).
  • Live Render Deployment: Connected to our registry hooks on Render for automated continuous deployment with zero downtime.

🔗 Public Links & Proofs

Everything is 100% open-source and publicly accessible without any auth walls:

🚀 Running it Locally

If you want to spin it up on your local machine with Google Gemma:


bash
# 1. Clone & install
git clone [https://github.com/GouravGC/tinysteps-ai.git](https://github.com/GouravGC/tinysteps-ai.git)
cd tinysteps-ai
pip install -r requirements.txt

# 2. Start local Ollama model (Google Gemma)
ollama run gemma3:4b

# 3. Launch Streamlit UI
python -m streamlit run app.py
Enter fullscreen mode Exit fullscreen mode

Top comments (0)