Enhancement: Hybrid Reasoning Step for Similarity Confirmation

[Kavirubc]

Problem

Currently, Simili Bot relies purely on vector similarity (mathematical distance) to identify duplicates. While efficient, this can lead to false positives where issues share similar keywords but have different technical root causes.

Proposed Solution

Implement a "Hybrid Reasoning Step" that uses an LLM to verify similarity results before taking action.

Process:

1. Discovery: Use the existing Vector Search to find the Top 5 most similar issues.
2. Reasoning: Send the new issue and the Top 5 candidates to an LLM.
3. Confirmation: The LLM analyzes the technical context and confirms if any are true duplicates.
4. Action: Only trigger the triage workflow (or the two-phase cooldown) if the LLM provides a high-confidence match.

Benefits:

• Precision: Combines the speed of vector search with the deep reasoning of an LLM.
• Context Awareness: Can ignore issues that look similar (e.g., both mention 'Auth') but describe different bugs.
• Better Triage: Enables more nuanced decisions, such as "Issue A is a duplicate of B, but also related to C."

Related Issues:

• Enhancement: Two-phase duplicate detection with cooldown period #54: Two-phase duplicate detection with cooldown period

[gh-simili-bot]

Simili Triage Report

Note

Quality Score: 8.5/10 (Good)
The issue is well-described.

Classification

Category	Value
Labels

Similar Threads

Similarity	Thread	Status
85%	#54 Enhancement: Two-phase duplicate detection with...	Open
83%	#55 Enhancement: Human activity detection for dupli...	Open
83%	#30 Similar Issue	Open

---

_{Generated by Simili Bot}