Parent

• Parent issue: Milestone 1 - Shardlake Post-Prototype Roadmap #5

Goal

Turn the current search service into a production-like query pipeline.

Detailed tasks

• 5.1 Query pipeline abstraction
  • Implement modular pipeline stages for:
    • query embedding
    • centroid routing
    • shard loading
    • ANN candidate search
    • merge
    • optional rerank
• 5.2 Parallel shard search
  • Execute shard searches concurrently.
  • Use the project async runtime.
• 5.3 Result merge logic
  • Combine shard-local results.
  • Maintain a global top-k heap.
• 5.4 Distance metrics
  • Support:
    • cosine
    • dot product
    • euclidean
• 5.5 Query configuration
  • Expose configurable parameters:
    • top_k
    • candidate_shards
    • rerank_limit
    • distance_metric

Definition of done

• Query execution is broken into explicit pipeline stages.
• Parallel shard search improves latency.
• Result merging and distance metrics are configurable and testable.

Child issue breakdown

• 5.1 - Build modular query pipeline stages #59 5.1 - Build modular query pipeline stages
• 5.2 - Add parallel shard search execution #60 5.2 - Add parallel shard search execution
• 5.3 - Implement global result merge logic #61 5.3 - Implement global result merge logic
• 5.4 - Add cosine, dot-product, and euclidean distance metrics #62 5.4 - Add cosine, dot-product, and euclidean distance metrics
• 5.5 - Expose query execution configuration #63 5.5 - Expose query execution configuration

Dependency summary

• 5.2 - Add parallel shard search execution #60 depends on 5.1 - Build modular query pipeline stages #59.
• 5.3 - Implement global result merge logic #61 depends on 5.1 - Build modular query pipeline stages #59.
• 5.4 - Add cosine, dot-product, and euclidean distance metrics #62 depends on 5.1 - Build modular query pipeline stages #59.
• 5.5 - Expose query execution configuration #63 depends on 5.2 - Add parallel shard search execution #60, 5.3 - Implement global result merge logic #61, and 5.4 - Add cosine, dot-product, and euclidean distance metrics #62.

Dependency graph

    #59
  /   |   \
#60  #61  #62
  \   |   /
    #63