Add similarity search functions for tasks and tickets

[dolliecoder]

This PR introduces SQL helper functions to enable vector similarity search for tasks and tickets using the pgvector setup added in PR1.

It is a follow-up, incremental step toward Issue #65. While PR1 introduced the description_embedding vector(768) columns for tasks and tickets (storage layer), this PR builds on that foundation by adding database-level similarity search functions (retrieval layer).
No embedding generation, indexing, or AI service integration is included here. This PR strictly enables semantic retrieval capability at the database level.

Dependency Note:
This PR depends on PR1, as it relies on the description_embedding columns introduced there. PR1 must be merged before this PR to ensure the functions execute against an existing schema. pr1 : #160

Changes Made

Added get_similar_tasks(query_embedding, match_count) SQL function
Added get_similar_tickets(query_embedding, match_count) SQL function

Each function:
Computes cosine similarity using <=>
Returns top-k most semantically similar rows
Ignores rows without embeddings (IS NOT NULL)
Added new Supabase migration file to maintain proper migration ordering

✅ Checklist

I have read the contributing guidelines.

I have added tests that prove my fix is effective or that my feature works.
(Not applicable – database-level capability addition only.)

I have added necessary documentation (if applicable).
(Not required at this stage.)

Any dependent changes have been merged and published in downstream modules.
(Depends on PR1 – embedding schema changes.)

[coderabbitai]

📝 Walkthrough

Walkthrough

This pull request adds two PostgreSQL PL/pgSQL functions for semantic similarity search. The get_similar_tasks and get_similar_tickets functions compute cosine similarity between query embeddings and stored embeddings, returning ranked results up to a configurable limit.

Changes

Cohort / File(s)	Summary
Database Vector Search Functions supabase/migrations/20251021110000_task_ticket_vector_search.sql	Adds two new PL/pgSQL functions (get_similar_tasks and get_similar_tickets) that perform semantic similarity search using vector embeddings. Both functions accept a query embedding (768-dimensional vector) and match count parameter, computing cosine similarity via the <=> operator and returning ranked results with similarity scores.

Estimated code review effort

🎯 2 (Simple) | ⏱️ ~10 minutes

Poem

🐰 Embeddings dance in vectors bright,
Eight seventy-four dimensions take flight,
Cosine whispers "find what's near,"
Similar tasks and tickets appear!
Search with similarity, clear and light. ✨

🚥 Pre-merge checks | ✅ 4

✅ Passed checks (4 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title 'Add similarity search functions for tasks and tickets' directly and clearly summarizes the main change—adding two SQL functions for vector-based similarity search on tasks and tickets.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.
Merge Conflict Detection	✅ Passed	✅ No merge conflicts detected when merging into main
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

✨ Finishing touches

• 📝 Generate docstrings

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Post copyable unit tests in a comment

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

[coderabbitai]

Actionable comments posted: 1

🤖 Fix all issues with AI agents

In `@supabase/migrations/20251021110000_task_ticket_vector_search.sql`:
- Around line 17-20: Add HNSW vector indexes so the <=> similarity ORDER BY on
description_embedding uses an index: create IF NOT EXISTS idx_tasks_embedding on
tasks using hnsw for column description_embedding with vector_cosine_ops, and
create IF NOT EXISTS idx_tickets_embedding on tickets similarly for
description_embedding; ensure the new indexes are applied before running
similarity queries (also add the same index when you see other queries ordering
by description_embedding <=> query_embedding).

🧹 Nitpick comments (1)

supabase/migrations/20251021110000_task_ticket_vector_search.sql (1)

1-23: Consider LANGUAGE sql and marking as STABLE.

Since the function body is a single RETURN QUERY SELECT, PL/pgSQL is unnecessary overhead — plain LANGUAGE sql avoids the PL/pgSQL executor layer. Additionally, these functions have no side effects and return consistent results for the same inputs within a transaction, so marking them STABLE lets the planner optimize repeated calls.

♻️ Suggested diff

 CREATE OR REPLACE FUNCTION get_similar_tasks(
     query_embedding vector(768),
     match_count INT DEFAULT 3
 )
 RETURNS TABLE (
     task_id UUID,
     title TEXT,
     description TEXT,
     similarity FLOAT
-) AS $$
-BEGIN
-    RETURN QUERY
+) LANGUAGE sql STABLE AS $$
     SELECT
         t.id AS task_id,
         t.title,
         t.description,
         1 - (t.description_embedding <=> query_embedding) AS similarity
     FROM tasks t
     WHERE t.description_embedding IS NOT NULL
     ORDER BY t.description_embedding <=> query_embedding
     LIMIT match_count;
-END;
-$$ LANGUAGE plpgsql;
+$$;

Same applies to get_similar_tickets below.

[dolliecoder]

Just checking in on this PR I will be happy to revise based on reviews. Please let me know if there’s anything I can adjust to help in moving this forward.