$199.00
GPU PM
Choose a search model tested on 30 real queries
For developers building document search or RAG who need to choose an embedding model using their own content. I compare three supported models on your documents and 30 real search queries. You receive measured retrieval results, examples of missed documents and a runnable search test. Choose based on whether the right information is found, rather than a general model ranking.
What you get
- A comparison of three supported embedding models, including open models where suitable, using one shared retrieval setup.
- One corpus of up to 100 plain-text documents, 250,000 characters in total, tested against 30 queries with your relevance labels.
- A simple keyword-search baseline, so you can see whether embeddings improve these searches.
- The share of queries with a relevant document in the first five results, plus query-by-query results and missed documents.
- Observed indexing/query times and embedding usage costs in the tested environment.
- A short recommendation, raw rankings, CSV results and Python scripts to rebuild the index and rerun the comparison.
- Up to $10 of test API usage included. One clarification round within seven calendar days of delivery.
Scope: one language, one fixed text-splitting method and one retrieval configuration. Results measure retrieval, not the quality of a chatbot’s generated answers.
Contact us before ordering if your corpus needs extraction or custom infrastructure.
Output format
PDF, CSV, JSON, Python ZIP
What to include
- Up to 100 plain-text documents in TXT, CSV or JSON, totaling no more than 250,000 characters, with stable document IDs.
- 30 realistic search queries and the IDs of the documents that should be relevant for each query.
- Your intended use and examples of searches that currently fail.
- Three candidate embedding models, or permission for me to choose supported options within the test allowance.
- Confirmation that the supplied content may be processed by the selected hosted providers. Remove confidential or personal information before sharing.
Provide a readable file link, or paste a small dataset into your request. Do not send passwords or API keys. Testing starts once the corpus, queries and relevance labels are complete.
Expected completion
4 business days
Not included
PDF/OCR extraction, creating relevance labels, scraping, multilingual evaluation, custom model installation, fine-tuning, reranker comparisons, repeated chunk-size optimization, production vector-database setup, chatbot development or generated-answer evaluation. No guarantee that one model will win or that results generalize beyond the supplied corpus and queries.