Devlog #1: MVP Scope and Classification Pipeline

Overview

Jetson is an autopilot for customer feedback that reads Help Scout inboxes and automatically converts conversations into bugs, feature requests, and documentation drafts synced with GitHub.

This week we focused on MVP scope decisions and building the core classification pipeline.

Key Decisions This Week

MVP Scope

We prioritized shipping the classifier and GitHub integration first, deferring documentation auto-publishing to version 2 to mitigate accuracy concerns.

Classification Accuracy

Current baseline: 77% accuracy with a target of 90%, measured through weekly precision and recall checks across categories.

Deduplication Strategy

The system implements embeddings-based clustering to consolidate duplicate issues rather than creating multiple copies.

GitHub Integration

Issues closed in GitHub automatically mark as resolved in Jetson, with project column mapping for non-technical team members.

Closing the Loop

When issues ship, Jetson identifies reporters and drafts bulk replies maintaining team voice.

What Was Built

• Classification pipeline connecting Help Scout webhooks to GitHub issue creation
• Embeddings-based deduplication clustering similar issues
• Mobile-responsive interface for approvals and duplicate merging

Real-World Test Results

During Recipe Kit’s React migration, over 45 support emails arrived within hours. Jetson grouped similar bug reports and created unified GitHub issues with user counts.

Manual triage typically requires about 4 hours; with Jetson, the process took approximately 15 minutes.

Progress & Next Steps

MVP is 75% complete. Upcoming work includes:

• Improving accuracy to 90%
• Building GitHub project sync
• Adding Help Scout webhooks
• Rebuilding UI with Shadcn components