Back
Year
2026
Tech Stack
Python, FastAPI, PostgreSQL (Neon), SQLAlchemy, Alembic, Groq LLM, instructor, pytest, GitHub Actions, Render
Description
An LLM-powered productivity agent that converts natural-language
WhatsApp messages into tracked and prioritised tasks. The LLM is responsible for
language understanding and intent extraction, while deterministic Python logic
controls priority scoring, date resolution, scheduling, and task state.
Why This Project Matters
Most AI agent systems allow the LLM to make operational decisions
directly, which can make behaviour unpredictable and difficult to test. Edith
demonstrates an alternative architecture where the LLM handles language while
deterministic Python logic owns task decisions, making the system more
predictable, testable, and production-oriented.
Technical Highlights
- Achieved 95.2% intent accuracy in a Groq-based evaluation run
- Built a 431-test pytest suite covering the application logic
- Deployed the FastAPI service on Render with Neon PostgreSQL for persistent task storage
- Implemented a GitHub Actions CI/CD pipeline for automated testing and deployment
Key Features
- Natural-language task capture through WhatsApp
- Automatic task prioritisation
- Persistent task tracking with PostgreSQL
- Deterministic scheduling and task-state management
- Automated deployment pipeline through GitHub Actions
My Role
- 🧩 Worked as the sole developer across all four implementation phases
- 🗄️ Designed the PostgreSQL schema and managed database migrations with Alembic
- 🔗 Built the WhatsApp webhook handling and intent-extraction pipeline
- 🧮 Implemented deterministic priority-scoring and date-resolution logic
- 🧪 Built the 431-test suite, configured CI/CD, and deployed the service on Render with Neon PostgreSQL