An AI-powered automation tool that learns web-based data entry processes by observation and executes them autonomously, completely eliminating the need for hardcoded RPA rules or traditional EDI setups.

Moving data between modern web portals (e.g., retail purchase orders) and internal systems is highly manual. Traditional EDI integrations take months per portal. This solution bridges the gap using an LLM-backed Playwright agent that watches a user perform the task once, deduces the data mapping, and takes over — working with any origin website.

chicali-code/
├── app.py                    # FastAPI orchestrator (dashboard + API endpoints)
├── llm_engine.py             # Gemini API module with token tracking & cost metrics
├── main.py                   # Standalone CLI orchestrator (original PoC)
├── destination_mock.html     # Mock ERP form (System B)
├── requirements.txt          # Python dependencies
├── .env.example              # Environment variable template
├── .gitignore
├── templates/
│   └── dashboard.html        # Dashboard UI (KPIs, charts, records table, logs)
└── static/
    └── app.js                # Frontend logic (WebSocket, Chart.js, table rendering)

• Launches a Playwright browser with two tabs.
• Tab 1 (Origin): Any user-specified URL (configurable from the dashboard).
• Tab 2 (Destination): Local ERP form with intentionally different field naming.
• Injects JavaScript event recorders into both tabs.

• User demonstrates the data-entry process manually (copy from Tab 1 → paste into Tab 2).
• JavaScript captures all copy, click, input, and paste events with timestamps.
• Interaction logs + destination form schema are sent to Gemini.
• LLM infers a semantic field mapping dynamically (no hardcoded mappings).

• User navigates to a new page in Tab 1 (new product, new order, etc.).
• A generic DOM scraper extracts all visible text, inputs, tables, and data attributes.
• LLM transforms the scraped data using the learned mapping into a fill payload.
• Playwright autonomously fills the destination form.

Access at http://localhost:8000 after starting the server.

• Dynamic Origin URL — paste any website URL, no code changes needed.
• KPI Row — Total Tokens Consumed, Estimated Cost ($), Successful Mappings.
• Token Usage Chart — Line chart showing input/output tokens per LLM call.
• Mapping Accuracy Chart — Doughnut chart of successful vs. failed fills.
• Records Table — Spreadsheet-style log of all mapped fields and filled values.
• Live Agent Logs — Real-time scrolling terminal via WebSocket.

1. Clone and create a virtual environment:

   python -m venv .venv
   source .venv/bin/activate  # On Windows: .venv\Scripts\activate

2. Install dependencies:

   pip install -r requirements.txt
   playwright install chromium

3. Configure environment:

   cp .env.example .env
   # Edit .env with your Gemini API key

4. Run the dashboard:

   python app.py

   Open http://localhost:8000 in your browser.

5. Or run the standalone CLI version:

   python main.py

The agent works with any website. Tested with: