Tier 2 · fixed-price engagement
RAG system diagnostic
Why your retrieval returns the wrong passage, isolated to the stage that causes it, with each candidate fix measured before it is recommended.
Who this is for
A team whose retrieval system works in a demo and fails on real questions, with no way to tell which stage is at fault.
Fixed price
$5,500
- Duration
- 1 to 2 weeks
- Deposit
- $1,000
- Other currencies
- €4,990 · £4,250 · S$7,400
- Catalog
- T2-07
Thirty minutes with the engineer who would run the work, before any paperwork.
The work, stage by stage
In order, with what happens at each stage and what it produces. The timings are the ones the catalog prices, not an estimate written for this page.
Collect the failures that matter
Days 1 to 2
Real queries that failed, from logs or from the people complaining, rather than synthetic examples. The diagnostic is only as good as the failure set it works from.
Instrument the pipeline stage by stage
Days 3 to 4
Chunking, embedding, retrieval, reranking and context assembly examined separately. This is the step most teams have never taken, because it means looking at what each stage returned rather than only at what came out of the end.
Find the stage that fails first
Week 1 to 2
For each failing query, whether the right passage was ever retrieved at all. If it was not, nothing downstream could have rescued it, and reranking or prompt work would have been wasted effort.
Test the fix before recommending it
Week 2
A candidate change, chunk size, embedding model, hybrid search, reranking or query rewriting, is run against the failure set, so the recommendation carries a measured before and after rather than an opinion.
Write it up by stage
On delivery
Findings ordered by the stage they belong to, with the measured effect of every change tested.
Everything that arrives, listed
The contents of the document, and the artifacts that come with it. If something here would not be useful to you, say so before we start and the scope changes.
- The failure set, each query attributed to the stage that broke it
- Per-stage findings for chunking, embedding, retrieval, reranking and context assembly
- A measured before and after for each change tested, so the recommendation is evidenced rather than asserted
- The retrieval configuration we would run, stated concretely enough to implement
- The instrumentation added during the diagnostic, left in place so the next regression is visible
What this covers, and who takes the rest
Stated up front, because scope is what derails an engagement in week two rather than at the proposal.
Inside this engagement
- The retrieval pipeline end to end, chunking through context assembly
- The real failing queries rather than synthetic ones
- Configuration changes, measured against the failure set
Handled separately
Named so you can plan and budget for it, whether we do it or somebody else does.
- Implementing the recommended configuration, which is a build engagement
- The generation layer's groundedness, which the hallucination and grounding audit measures
Start with a call, not a proposal.
Thirty minutes, straight to the engineer who would run the work. We will tell you if a smaller engagement answers your question, or if nothing here does.
Typical reply within one business day · Engagements start at $2,500
Fixed price · 1 to 2 weeks