Services

Nine engagements. One discipline: measure before you change anything.

Retridge works on retrieval-augmented and agentic systems over private data. Every engagement is scoped in writing, priced up front, and delivered against evidence you can re-run yourself.

Flagship engagement

The RAG System Audit

A two-week diagnostic of an existing AI or RAG system: structured evaluation on 50–100 real queries, failure classification by root cause, cost analysis, and a written report with a prioritized remediation roadmap.

Format

Fixed price

Indicative

$4,000–$7,500

Duration

2 weeks

Request a RAG Audit See the full process

What is evaluated

Retrieval quality recall@k · precision@k · ranking

Chunking & embeddings boundaries · size · model fit

Grounded generation faithfulness · completeness · citations

Data quality duplicates · staleness · structure

Security surface injection · leakage · access

Orchestration prompts · routing · tools

Cost & latency tokens · p50/p95 · cost per answer

Audit process

Five steps, ten working days.

01 Day 1–2

Kickoff

Architecture review, access setup, and agreement on the query set and what a correct answer looks like.

02 Day 3–6

Evaluation run

50–100 real queries executed with retrieval traces captured and scored per stage, not just end-to-end.

03 Day 5–8

Pipeline review

Ingestion, indexing, prompts, orchestration, security probes, latency and token accounting.

04 Day 8–10

Report

Scorecard, failure distribution, root-cause findings, and a remediation backlog ranked by impact and effort.

05 Day 10

Walkthrough call

A working session with your engineers. Every finding is traceable to a query, a chunk, and a number.

What we need from you

Read access to the system, or exports of the corpus

50–100 example queries, or production logs we can sample

Roughly 2 hours of stakeholder time across the two weeks

A named engineering contact for architecture questions

What you get back

Written report with executive assessment and technical detail

The evaluation dataset and harness — yours to keep and re-run

Prioritized remediation backlog, ready for your sprint board

A walkthrough session, plus 30 days of follow-up questions

Full catalog

The other eight engagements

Prices are indicative and confirmed in a fixed-scope proposal after the discovery call. Expand any engagement for scope and deliverables.

AI Readiness & Strategy Assessment

Fixed price · $3,000–$6,000 Entry point without an AI system

Evaluation of where AI creates value in your operations, a data readiness review, build-vs-buy guidance, and a phased adoption roadmap. For organizations that do not yet have an AI system to audit.

Deliverables

Opportunity map by function Data readiness review Build-vs-buy recommendation Phased adoption roadmap

AI Security & Red-Team Assessment

Fixed price · $6,000–$12,000 For customer-facing AI

Adversarial testing of a deployed AI system: prompt injection, jailbreaks, data leakage, access-control bypass, and misuse scenarios. Findings are risk-classified and paired with a hardening plan your team can execute.

Test classes

Direct & indirect prompt injection Retrieval-scope and permission bypass Data exfiltration via context Tool and agent misuse paths

Production RAG Build

Project · $10,000–$30,000 Scoped by data and integration

End-to-end design and implementation of a retrieval-augmented assistant over your private data, with an evaluation pipeline, observability, and handover documentation. Built so your team can own it after we leave.

Included

Ingestion and indexing pipeline Hybrid retrieval and reranking Evaluation harness and CI checks Tracing, observability, handover docs

Agentic Workflow Design

Project · from $15,000 Premium / advanced

Retrieval-aware agent systems that plan, retrieve, and act across multistep tasks. We design the control flow, tool integrations, iteration strategy, and the tests that keep an agent from quietly going wrong at step four.

Focus

Control-flow and planning design Tool integration and permissions Iteration and termination strategy Step-level testing and tracing

Knowledge-Graph RAG

Project · scoped per engagement Differentiator service

Graph-based retrieval (e.g. Neo4j) for highly connected data — legal, supply chain, finance — including query decomposition and multihop retrieval. We will also tell you when a graph is not justified and vector search is enough.

Suited to

Multihop and comparative questions Entity-dense contract or claim corpora Supply chain and dependency queries Hybrid graph + vector retrieval

Optimization & Reliability Retainer

Retainer · $2,000–$5,000/month Post-launch ownership

Ongoing tuning of accuracy, latency, and token cost; regression evaluations before releases; and a monthly performance report. The point is that a model upgrade should never be a surprise.

Cadence

Regression evals gating releases Monthly performance report Cost and latency tuning Eval dataset maintenance

Corporate Training & Workshops

$5,000–$15,000 per workshop 1–2 days · onsite or remote

One- to two-day workshops for engineering and product teams: RAG design, evaluation discipline, failure diagnosis, AI security awareness, and production readiness. Half-day executive briefings are also offered. See the Training page.

Modules

LLM behavior & failure modes Retrieval design Evaluation-first development Security, misuse & production tuning

Fractional AI Engineering

Monthly retainer · scoped 10–20 hours per week

Ten to twenty hours a week embedded with one client as their AI engineering lead — for mid-size companies not ready to hire full-time AI talent. Only one fractional client is taken at a time.

Typical remit

Technical direction and architecture Hiring and vendor evaluation support Hands-on implementation review Team mentoring

FAQ

Questions we get before the first call

If yours isn't here, ask it on the discovery call. It's free and there is no pitch.

What if we don't have an AI system yet?

Start with the AI Readiness & Strategy Assessment. It establishes where retrieval actually helps, whether your data is ready, and what to build first — so you don't audit a system that shouldn't have been built.

What do you need from us?

Read access to the system or exports of the corpus, 50–100 example queries (or logs we can sample), about two hours of stakeholder time, and one named engineering contact. That's it.

What tools and stacks do you work with?

Python-based stacks, the major vector databases, hybrid search, rerankers, Neo4j for graph retrieval, and the mainstream LLM providers. We are deliberately not tied to one vendor — the audit reports what your stack does, not what we would have chosen.

Is our data safe during an audit?

We work under NDA, prefer read-only access scoped to what the evaluation needs, and can run in your environment. Sensitive corpora can be sampled or redacted; findings are reported without reproducing confidential content.

Do you build, or only advise?

Both. Many clients take the audit and implement the backlog themselves — that is a good outcome. When you'd rather we build it, the Production RAG Build and Agentic Workflow engagements pick up where the audit ends.

How is the price fixed if you don't know what you'll find?

Because the scope is the measurement, not the fix. The audit runs a defined number of queries across a defined set of layers. What we find changes the roadmap, not the invoice.

Not sure which engagement fits?

Describe the symptoms in 30 minutes and we'll tell you which one — or whether you need one at all.

Book a free 30-minute consultation