Back to Blog
ProductDecember 20248 min read

From 6.5 to 9.0: The AI Tickets Quality Journey

How we improved ticket generation quality by 38% in one month

When we launched AI Tickets in October, we knew we had something powerful. But we also knew it wasn't perfect. Tickets sometimes missed context, referenced non-existent components, or left gaps in implementation details.

So we set out on a mission: improve ticket quality from 6.5/10 to production-ready. Through iterative improvements over the following months, we achieved 9.0/10. Here's how.

The Starting Point

Our initial release worked, but quality assessments revealed gaps. Tickets scored 6.5/10 with common issues:

  • 4 gaps per ticket on average — missing implementation details
  • Generic code references — "ticketService" instead of actual file paths
  • Fabricated components — suggesting files that didn't exist
  • Vague task descriptions without evidence citations

We needed a systematic approach to improve quality. So we built one, version by version.

The Journey: Atomic v1.2 → Atomic v1.5

Atomic v1.3Advanced Taxonomy Features

We introduced taxonomy-based filtering to understand what type of feature was being requested. This allowed us to match requirements with the right code patterns.

8.5/10
Quality Score
0
Gaps
+31%
vs Atomic v1.2
Atomic v1.4Enhanced Context

We enriched prompts with rich metadata: API endpoints, function signatures, class structures, and tech stack information. The LLM now had much more context to work with.

+6.25%
Quality Improvement
+20%
Code Precision
+12.5%
Task Specificity
Atomic v1.5Integration Points Analysis

We added automatic detection of external dependencies: APIs, databases, cloud services, and third-party libraries. Tickets now include comprehensive integration context.

9.0/10
Quality Score
+∞
Dependencies (was empty)
+58%
Integration Awareness

Infrastructure That Powers It

Quality improvements weren't just about better prompts. We also rebuilt the foundation that makes AI Tickets possible.

Database Indexing

We redesigned our code indexing system with a comprehensive schema, enabling richer metadata extraction and more reliable code analysis.

99.78%
Success rate (up from 70%)

File Taxonomy

Enhanced classification system that understands file categories, frameworks, and architecture layers.

18x
Faster config file processing

The Results

6.5 → 9.0
Quality Score
+38% improvement
4 → ~0
Avg Gaps per Ticket
Significant reduction
99.78%
Indexing Success
Up from 70%

What this means for developers using AI Tickets:

  • Dramatically fewer gaps — tickets now include comprehensive implementation details
  • Actual code references — no more fabricated components or generic paths
  • Integration awareness — automatic detection of APIs, databases, and cloud services
  • Better task breakdown — more granular, actionable implementation steps

What's Next

We're not done. The journey from 6.5 to 9.0 taught us that systematic, iterative improvement works. We're continuing to refine AI Tickets with:

  • API endpoint prioritization for better context matching
  • Dependency analysis for more complete task breakdowns
  • Layer-aware completeness to prevent missing implementation steps
Try AI Tickets