Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Diagnoses and improves Qdrant search relevance. Use when someone reports 'search results are bad', 'wrong results', 'low precision', 'low recall', 'irrelevant matches', 'missing expected results', or asks 'how to improve search quality?', 'which embedding model?', 'should I use hybrid search?', 'should I use reranking?', 'how to measure retrieval quality?', 'build a golden set', 'ground truth dataset', or 'how to score recall@k?'. Also use when search quality degrades after quantization, model c
.claude/skills/qdrant-qdrant-search-quality/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-04 | ✗→✓ | ▲ Improved | 9% | 0% |
| case-21 | ✓→✓ | ＝ Same ✓ | -2% | 0% |
| case-03 | ✓→✓ | ＝ Same ✓ | -28% | 0% |
| case-05 | ✓→✓ | ＝ Same ✓ | 15% | 0% |
| case-06 | ✓→✓ | ＝ Same ✓ | 17% | 0% |
First determine whether the problem is the embedding model, Qdrant configuration, or the query strategy. Most quality issues come from the model or data, not from Qdrant itself. If search quality is low, inspect how chunks are being passed to Qdrant before tuning any parameters. Splitting mid-sentence can drop quality 30-40%.
Isolate the source of quality issues, establish labeled baselines to measure recall and relevance, tune HNSW parameters, and choose the right embedding model. Diagnosis and Tuning
Hybrid search, reranking, relevance feedback, and exploration APIs for improving result quality. Search Strategies
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-21 | pass→pass | 32,141 | 24,704 | -23% | 1 | 1 | 0% | 4,660 | 4,574 | -2% | 0 | 0 | — |
case-01 | fail→fail | 22,270 | 22,915 | +3% | 1 | 1 | 0% | 3,838 | 3,999 | +4% | 0 | 0 | — |
case-02 | fail→fail | 24,883 | 19,845 | -20% | 1 | 1 | 0% | 4,187 | 3,402 | -19% | 0 | 0 | — |
case-03 | pass→pass | 21,005 | 21,507 | +2% | 1 | 1 | 0% | 3,418 | 2,464 | -28% | 0 | 0 | — |
case-04 | fail→pass | 17,799 | 21,630 | +22% | 1 | 1 | 0% | 2,864 | 3,135 | +9% | 0 | 0 | — |
case-05 | pass→pass | 23,265 | 23,310 | +0% | 1 | 1 | 0% | 3,577 | 4,125 | +15% | 0 | 0 | — |
case-06 | pass→pass | 17,591 | 18,092 | +3% | 1 | 1 | 0% | 3,096 | 3,612 | +17% | 0 | 0 | — |
case-07 | pass→pass | 15,760 | 18,002 | +14% | 1 | 1 | 0% | 2,834 | 2,827 | -0% | 0 | 0 | — |
case-08 | pass→pass | 6,940 | 5,683 | -18% | 1 | 1 | 0% | 914 | 1,228 | +34% | 0 | 0 | — |
case-09 | pass→pass | 7,806 | 7,052 | -10% | 1 | 1 | 0% | 1,365 | 1,554 | +14% | 0 | 0 | — |
case-10 | pass→pass | 15,360 | 13,414 | -13% | 1 | 1 | 0% | 2,658 | 2,453 | -8% | 0 | 0 | — |
case-11 | pass→pass | 10,727 | 8,944 | -17% | 1 | 1 | 0% | 1,774 | 1,618 | -9% | 0 | 0 | — |
case-12 | pass→pass | 13,374 | 9,618 | -28% | 1 | 1 | 0% | 2,174 | 1,952 | -10% | 0 | 0 | — |
case-13 | pass→pass | 16,807 | 14,730 | -12% | 1 | 1 | 0% | 2,408 | 2,281 | -5% | 0 | 0 | — |
case-14 | pass→pass | 17,347 | 18,109 | +4% | 1 | 1 | 0% | 2,507 | 2,836 | +13% | 0 | 0 | — |
case-15 | pass→pass | 20,577 | 21,609 | +5% | 1 | 1 | 0% | 3,594 | 3,419 | -5% | 0 | 0 | — |
case-16 | pass→pass | 18,626 | 16,329 | -12% | 1 | 1 | 0% | 2,940 | 2,372 | -19% | 0 | 0 | — |
case-22 | pass→pass | 18,212 | 15,121 | -17% | 1 | 1 | 0% | 3,221 | 3,104 | -4% | 0 | 0 | — |
case-17 | pass→pass | 13,833 | 8,612 | -38% | 1 | 1 | 0% | 1,952 | 1,399 | -28% | 0 | 0 | — |
case-18 | pass→pass | 9,897 | 9,161 | -7% | 1 | 1 | 0% | 1,461 | 1,623 | +11% | 0 | 0 | — |
case-19 | pass→pass | 17,964 | 16,303 | -9% | 1 | 1 | 0% | 3,027 | 3,054 | +1% | 0 | 0 | — |
case-20 | pass→pass | 20,961 | 16,238 | -23% | 1 | 1 | 0% | 3,101 | 2,644 | -15% | 0 | 0 | — |
case-23 | pass→pass | 20,962 | 21,711 | +4% | 1 | 1 | 0% | 3,393 | 3,731 | +10% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 23 cases were attempted. The headline lift of +4 percentage points is the difference between those two pass rates over the 23 comparable cases.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.