A location-aware historical context application that brings history to life through AI-powered narratives. Users can explore the historical significance of any location by simply providing coordinates and a radius, then engage in an interactive conversation to learn more about the area's past.

One Click History is a full-stack application built using concept-oriented design principles. The backend exposes location-based historical context generation through a REST API, while the frontend (in a separate repository) provides an intuitive interface for users to explore history through their device's GPS.

• Location-Based Historical Narratives: Generate rich, contextual historical information for any geographic location
• Interactive Q&A: Ask follow-up questions grounded in the location's historical context
• Session Management: Save and resume conversations about specific locations
• User Authentication: Secure user accounts with unique profiles
• Chat History: Browse previous explorations without loading full conversation data

Built using concept-oriented design, the application is structured around three core concepts:

1. UserAuthentication: Manages user registration, login, and profile identification
2. AIHistoricalContextAgent: Generates location-aware historical narratives with built-in validation to prevent hallucinations
3. LocationChatLedger: Stores chat metadata as a reference system for browsing past conversations

• Backend: TypeScript with Deno runtime
• Database: MongoDB Atlas
• LLM: Google Gemini 2.5 Flash
• Framework: Custom concept engine with synchronizations
• Design Tool: Context (markdown-based LLM collaboration)

• Design Changes: Evolution of the concept architecture from Assignment 2 through Assignment 4b
• Reflection: Personal reflections on the development process, challenges, and insights about LLM-assisted software development
• Trace Log: Example server execution trace showing the system in action
• Final Video: Demonstration of the complete application (link to be added)

• Deno installed
• MongoDB Atlas account (free tier)
• Google Gemini API key

1. Clone the repository

git clone <repository-url>
cd one_click_history

2. Configure environment variables in .env:

GEMINI_API_KEY=your_key_here
GEMINI_MODEL=gemini-2.5-flash
MONGODB_URL=your_mongodb_connection_string
DB_NAME=your_database_name

3. Compile Context tool (optional):

deno compile -A --output ctx .ctx/context.ts

4. Run the server:

deno task concepts

The API server will start on http://localhost:8000

Complete API documentation is available in design/API-DOCUMENTATION.md, including:

• Endpoint specifications for each concept
• Request/response formats
• Example workflows with curl commands
• Error handling details

Individual concept specifications:

• UserAuthentication API
• LocationChatLedger API
• AIHistoricalContextAgent API

Run all tests:

deno test -A

This project was developed using the Context tool, a markdown-based framework for design-driven development with LLM collaboration. All design decisions, iterations, and LLM interactions are preserved in the context/ directory.

Key design artifacts:

• design/background/: Concept design principles and implementation guides
• design/concepts/: Individual concept specifications and implementation notes
• context/: Complete version history of all design documents

src/
├── concepts/           # Core concept implementations
│   ├── UserAuthentication/
│   ├── AIHistoricalContextAgent/
│   └── LocationChatLedger/
├── syncs/             # Synchronization rules between concepts
├── engine/            # Concept engine framework
└── main.ts            # Application entry point

design/
├── concepts/          # Concept specifications
├── background/        # Design principles and guides
└── API-DOCUMENTATION.md

context/               # Complete design history (preserved)

The AIHistoricalContextAgent implements a sophisticated validation system:

• Secondary checker agent validates location claims before accepting responses
• Hemisphere verification prevents geographic mismatches
• Retry mechanism with feedback allows the LLM to correct mistakes
• Up to 3 attempts with cumulative validation feedback

• LocationChatLedger stores only metadata (location, timestamp, main location name)
• AIHistoricalContextAgent manages full conversation history and state
• This separation enables efficient chat browsing without loading entire conversation histories

Developed for MIT 6.104: Software Studio (Fall 2025)

Assignment series exploring concept-oriented design, LLM-assisted development, and full-stack application architecture.

Davis Lee