ML Advisor

A mobile app that helps software engineering students and developers choose a machine-learning model for software bug prediction. It combines a model library, side-by-side comparison, a guided recommendation wizard and an AI chat grounded in published research.

Capstone project · API: https://ml-advisor-api.onrender.com

Team project. Built by a team of four. See Team for who did what.

Screenshots:

Features

• Email and Google sign-in (Firebase Authentication) with user roles: student, developer and admin.
• Model library and detail pages with performance information drawn from published research.
• Comparison view with bar charts (fl_chart).
• Recommendation wizard: four questions (dataset size, dataset type, priority, class imbalance) feed a weighted scoring routine that ranks models.
• Favourites, research papers and a glossary.
• AI chat: a FastAPI endpoint matches your question against a curated knowledge base of research summaries and passes the best matches to an LLM.
• Admin dashboard: create, update and delete models, papers, glossary terms and users, with usage analytics.

Architecture

flowchart LR
  A[Flutter app] -->|Auth, models, papers, glossary, favourites| B[(Firebase Auth + Firestore)]
  A -->|POST /chat| C[FastAPI service on Render]
  C --> D[Keyword retrieval over research knowledge base]
  D --> E[OpenRouter: Llama 3.2 3B Instruct]
  E --> C --> A

Loading

The retrieval step is keyword-based: it selects relevant research summaries from backend/knowledge_base/ and adds them to the prompt. It does not use embeddings or a vector database. The knowledge base was built from 26 peer-reviewed papers.

Tech stack

Area	Technology
Mobile app	Flutter, Dart, Provider, fl_chart
Auth and data	Firebase Authentication, Cloud Firestore
Backend	Python, FastAPI, httpx, python-dotenv
LLM	Llama 3.2 3B Instruct via OpenRouter
Hosting	Render (API)

Repository layout

backend/
  main.py                  FastAPI app with the /chat endpoint
  knowledge_base/          Research summaries and retrieval logic
  requirements.txt
mobile/ml_advisor_app/     Flutter app (lib/screens, providers, services, models)

Getting started

Prerequisites: Flutter SDK (Dart >=3.0 <4.0), Python 3.10+, an OpenRouter API key, a Firebase project.

1. Backend

cd backend
python -m venv .venv
source .venv/bin/activate        # Windows: .venv\Scripts\activate
pip install -r requirements.txt
echo "OPENROUTER_API_KEY=<your-key>" > .env
uvicorn main:app --reload --port 8000

Run it from the backend/ folder so the knowledge_base package is found.

2. Mobile app

cd mobile/ml_advisor_app
flutter pub get

By default the app calls the hosted API. To use your local backend, edit lib/utils/constants.dart:

• Android emulator: http://10.0.2.2:8000
• iOS simulator: http://localhost:8000
• Physical device: your computer's LAN address, port 8000

Then run:

flutter run

3. Firebase

The repository contains the Firebase client configuration for the original project. To use your own project, run flutterfire configure, enable Email/Password and Google sign-in, and create the Firestore collections the app reads: models, papers, glossary, favorites, users and analytics_history.

Seed data for these collections is not included in this repository yet.

4. Build an Android APK

cd mobile/ml_advisor_app
flutter build apk --release

The APK is written to build/app/outputs/flutter-apk/app-release.apk. The release build currently uses the debug signing key (android/app/build.gradle.kts), which is fine for sideloading and demos but not for the Play Store. A published APK is not available yet.

Testing

The Flutter project still has the default template widget test. Automated tests for the recommendation logic and the chat endpoint are planned.

Roadmap

• Authentication and rate limiting for /chat; stop returning raw error text to clients
• Move the recommendation weights into configuration or Firestore
• Seed script and Firestore security rules in the repository
• Unit tests for the scoring routine and the knowledge-base retriever
• Embedding-based retrieval

Team

Member	Contribution
Shine Chikwapulo	Team lead; backend, knowledge base and AI chat; most of the Flutter implementation (per commit history); set up the Git workflow
Henno	Frontend
Jayden	Documentation
Dube	Documentation

Author

Shine Chikwapulo · GitHub · LinkedIn