Work Tool · AI
AI Ticket Triage & Insight Dashboard
Screenshot shows sanitized example data, not real customer or employee information.
Internal dashboard for Unity's Partner Relations team to self-serve open tickets and years of historical customer engagement notes, replacing manual searching with fast, filterable views. Powered by a resumable async batch pipeline using the Anthropic SDK to distill thousands of raw records into concise, LLM-generated summaries.
Architecture
A static frontend paired with an offline batch pipeline: Python scripts pull raw ticket and legacy note data, summarize it asynchronously through the Anthropic SDK, and cache the results as compressed JSON for the browser to filter instantly, with no database or backend server.
Step 1
Raw data export
Ticket and legacy note data is exported to CSV/JSON on a scheduled basis.
Step 2
Async batch summarization
A Python script using an async Anthropic client with bounded concurrency processes each record, skipping ones already summarized so re-runs stay cheap.
Step 3
Compressed JSON cache
Summaries are written as gzip-compressed per-account JSON files, avoiding the need for a database.
Step 4
Static dashboard
A dependency-free HTML/JS frontend loads the cached data and handles filtering, search, and person lookup entirely client-side.
Key decisions
- Chose a resumable batch job over live per-request summarization, so the tool stays usable without waiting on LLM latency for every page load.
- Skipped a database entirely: the dataset is small enough that gzip JSON outperforms query overhead, and a static site is trivial to host.
- Added a live on-demand summarization action for the small number of records that need an update between batch runs.