top k
technique judged 93% likely a named technique or architecture
See how it has spread across media tiers
Mentions over time
Table view
| Week | Articles | Sources |
|---|---|---|
| 2024-W26 | 1 | 1 |
| 2024-W38 | 1 | 1 |
| 2024-W41 | 1 | 1 |
| 2026-W09 | 2 | 1 |
| 2026-W11 | 4 | 1 |
| 2026-W13 | 1 | 1 |
| 2026-W16 | 1 | 1 |
| 2026-W18 | 1 | 1 |
| 2026-W24 | 1 | 1 |
| 2026-W31 | 1 | 1 |
| 2026-W32 | 2 | 2 |
| 2026-W35 | 1 | 1 |
| 2026-W37 | 2 | 2 |
| 2026-W38 | 1 | 1 |
| 2026-W40 | 6 | 2 |
| 2026-W41 | 1 | 1 |
In their words
Architecture Overview ┌─────────────┐ ┌─────────────┐ ┌─────────────┐ ┌─────────────┐ │ Ingest │────▶│ Embedding │────▶│ Retrieval │────▶│ Generation │ │ (Documents)│ │ (Vectors) │ │ (Top-K) │ │ (Answer) │ └─────────────┘ └─────────────┘ └─────────────┘ └─────────────┘ │ │ │ │ ▼ ▼ ▼ ▼ Chunking Vector DB Reranking LLM + Context Strategy (Qdrant/ (Cross- (Citations, (Semantic/ Pinecone/ encoder/ Gua
flowchart LR Q[question] --> E{exact cache} E -- hit --> A[answer] E -- miss --> S{semantic cache} S -- hit --> A S -- miss --> R[embed + retrieve top-k<br/>optional retrieval cache] R --> P[prompt: instruction, passages, question<br/>optional provider prompt cache] P --> M[(LLM)] M --> A Loading Demo: replay 201 real requests and see which layer answered, the latency and the cost, for each cache
Claude Haiku 5.5 API’ye hızlı bakış Parametre Haiku 5.5 davranışı Model kimliği claude-haiku-5-5 — Bedrock: anthropic.claude-haiku-5-5 ; ayrı takma ad yok 100K token’a kadar istemler için MTok fiyatı 0,10$ giriş, 0,50$ çıkış, 0,01$ önbellek okuma 100K token üzeri istemler için MTok fiyatı 0,50$ giriş, 2,50$ çıkış, 0,05$ önbellek okuma Bağlam / maksimum çıkış 1M / 128K; toplu işlemde output-300k-20
Searching globally and filtering only the top-k would be a different architecture: other tenants' documents could consume the result budget, as well as expanding the opportunity for leakage.
Recall@k measures retrieval of predefined relevant evidence in the top k results.
Top-k retrieval can confidently return the wrong neighborhood.
At inference time, query the store with the user prompt, retrieve the top k snippets, and prepend them to the model prompt.
In the example below, $fetch_count controls the number requested from SEARCH, while $top_k controls the number returned to the application.