Overview

Logger is a ground-up rebuild of Chatable-Study-Database — a Python/Jupyter-based study time analysis system with basic LLM chat. The previous version relied on manual Excel logging, Jupyter notebooks for visualization, and a RAG pipeline for querying history.

The problem: Manual time tracking in spreadsheets is tedious, error-prone, and disconnects the act of logging from the insight it's meant to produce. You end up with 26 CSV files spanning three years of data, but no easy way to see patterns, track project evolution, or ask questions about your own productivity.

Logger solves this by combining real-time timer-based logging with an analytics dashboard, AI-powered natural language querying, and a project progression timeline — all in a polished, professional web interface that makes the daily act of tracking feel effortless.

Motivation

Productivity tracking shouldn't require a data science workflow. The original system worked — three years of continuous logging prove that — but each quarter required manual CSV creation, column formatting, and Jupyter notebook reruns just to see basic trends. Text descriptions were separated from time data. Category names drifted across quarters. Getting a simple answer like "how much time did I spend on research in 2024?" meant writing pandas queries.

Logger turns passive data collection into an active feedback loop: log time with timers, see patterns instantly in dashboards, and ask questions in natural language. The goal is to make productivity tracking a tool you actually use daily, not a chore you maintain weekly.

The project also serves as a case study in structured AI-assisted development using Propel — a constraint framework for Claude Code that enforces investigation-first methodology, human-in-the-loop gates, and domain-specific auditing throughout the development process.

Core Features

Architecture

Logger follows a clean client-server architecture with a SvelteKit frontend communicating with a FastAPI backend over REST APIs. The backend manages all data persistence, CSV import logic, timer state, analytics aggregation, and Claude API calls.

The frontend is organized into seven pages matching the sidebar navigation, each with dedicated components and stores. The backend uses a service-layer pattern with routers for HTTP routing and services for business logic.

Data Model

Logger's data model is session-centric. Each academic quarter (e.g., "Winter 2025") is a session containing categories, timer entries, manual entries, and daily observations. Categories are linked across sessions through category families, enabling project-level tracking over time.

Legacy Data

Logger imports 26 CSV files spanning Fall 2022 through Summer 2025, handling format variations (missing week/day columns in 2022, BOM characters, inconsistent column names in text CSVs) and normalizing category names with quarter suffixes into linked families.

AI Integration

Logger uses the Claude API for two features, both requiring explicit user approval in the UI before any LLM call is made:

Tech Stack

Why These Choices

Getting Started

git clone https://github.com/KevinBian107/logger.git
cd logger

cd backend
uv venv && source .venv/bin/activate
uv pip install -e .
uvicorn logger.main:app --reload

cd frontend
pnpm install
pnpm dev

# Place CSV files in data/ directory
# Use the Data page to import sessions

Application Pages

Logger is organized into seven pages, each accessible from the persistent sidebar navigation.

Acknowledgments

• Chatable-Study-Database — The original Python/Jupyter productivity tracking system that Logger rebuilds and extends.
• Propel — Structured constraint framework for Claude Code used throughout Logger's development, enforcing investigation-first methodology and human-in-the-loop gates.
• Code Manual — Documentation and methodology guides for AI-assisted development workflows.