Skip to content
RETRIDGE
How it works

From first call to delivered report in about three weeks.

No discovery phase that bills for itself, no open-ended statement of work. You know the price and the deliverables before the work starts, and you keep everything we build to measure your system.

Engagement timeline

Six stages. Two decision points.

You can stop after the report — most of the value is there, and plenty of clients implement the backlog themselves. Implementation is a separate decision, priced separately.

Start with an audit
  1. 01

    Discovery call

    Free · 30 min

    We look at the symptoms: what users ask, where answers go wrong, what the architecture looks like, what has already been tried. At the end of it you get an honest answer on whether an audit is worth your money.

    Output: a yes/no recommendation, and a rough shape of scope.
  2. 02

    Fixed-scope proposal

    2 working days

    A written scope: which layers are evaluated, how many queries, what access is needed, what you receive, the fixed price, and the dates. Nothing in it is hourly.

    Decision point: sign, adjust scope, or walk away at no cost.
  3. 03

    Audit

    Weeks 1–2

    Kickoff, evaluation run across 50–100 real queries, pipeline review from ingestion to production, security probes, and cost/latency accounting. Your engineers get a short written update mid-way, not silence.

    Your time required: roughly two hours, total.
  4. 04

    Report

    Day 10

    Executive assessment, scorecard, failure distribution, root-cause findings across the five reliability layers, and a remediation backlog ranked by user impact against engineering effort.

    Output: written report, evaluation dataset, and the eval harness.
  5. 05

    Walkthrough

    90 min

    A working session with your engineers. We open the traces, argue about the findings, and agree what the first three fixes should be. Followed by 30 days of follow-up questions at no charge.

    Decision point: implement in-house, or with us.
  6. 06

    Implementation

    Optional

    If you want the fixes built, the build engagement starts from the backlog and is measured against the same evaluation set — so improvement is a number, not a claim.

Checklist

What we need from you

Three things, and about two hours of your team's time. If you don't have a query set yet, we can build one from production logs — or from a workshop with the people who use the system.

Not ready to talk yet?

Download the one-page RAG Audit Checklist and run the first pass yourself.

Get the checklist →
System access, or exports
Read-only access to the assistant and its index, or exports of the corpus. A staging environment is fine — often better.
50–100 example queries, or logs
Real questions from real users, including the ones that embarrass the system. Logs we can sample work just as well.
Two hours of stakeholder time
Kickoff, one architecture conversation, and the walkthrough. Spread across two weeks.
Helpful, not required
Existing eval results, tracing data, an architecture diagram, and any list of known complaints.

Start with the call. It costs nothing and takes 30 minutes.

Book a free 30-minute consultation