Skip to content
kristenmartino.ai
Independent product and AI review

Find where your AI feature needs work.

I review its inputs, outputs and failure handling, then give your team a written assessment with examples and recommended fixes.

See the engagementsView project examples
Example deliverableoutline

Assessment outline.

01Observed issue and affected workflow
02Supporting example and testing limits
03Recommended change and priority
Findings · evidence · next steps
§ 01Start Here

Where would you start?

Choose the situation closest to yours to highlight a suggested engagement.

§ 02Engagements

Two review scopes.

Each includes a written assessment. We agree on the questions, access and scope before work begins.

01
~1 weekMostly async

Reality Check

from $1,500

For a team that wants a focused review of one existing AI feature.

What I do
  • Review one feature's inputs, outputs and user workflow
  • Exercise agreed failure cases and document examples
  • Check how users verify or correct the output
  • Rank recommended fixes by impact and effort
Deliverable

Feature assessment

  • —Findings with supporting examples
  • —Observed failures and likely causes
  • —Prioritized fixes
  • —Open questions and testing limits
Discuss a Reality Check
02
~2 weeksInterviews + async

Full Diagnosis + Roadmap

from $6,000

For a team deciding what to build or change across a product workflow.

What I do
  • Interview stakeholders and map the current workflow
  • Document requirements, constraints and unresolved questions
  • Compare options against agreed criteria
  • Sequence the recommended work and dependencies
Deliverable

Assessment and roadmap

  • —Problem statement and supporting evidence
  • —Options and tradeoffs
  • —Proposed implementation sequence
  • —Summary for the decision owner
Discuss a Full Diagnosis

Scope and price are agreed before work begins. Additional work is quoted separately.

§ 03Project examples

Related projects.

These independent projects show the kind of implementation and review work I do. Each case study states what is built, tested or still pending.

Read a sample assessment →

A self-initiated review of GTM Healthcare Intelligence, with findings, recommended fixes and supporting evidence.

  • Eval Harness

    A measured development baseline.

    Llama 3.1 8B matched 319/400 existing Sift category labels. The references are model-assigned; human label review and the final held-out model comparison remain pending.

    79.8% stored-label agreement · development set
    Read the case study →View development results ↗
  • Cabana

    AI triage and separate staff approval.

    AI can route a repair to a fixed $75 Stripe test deposit before staff approval. Staff approve work and assign visit times. One failure-injection run produced 50 Airtable records for 50 bookings with no duplicates.

    Recorded test: 50/50 Airtable records · no duplicates
    Read the case study →Try the demo ↗
  • Tarazu

    Relative scores and saved decision reasons.

    The fictional Relay walkthrough links editable scores, supporting and conflicting notes, and a draft decision. Users can copy the reasoning into a decision brief.

    Fictional walkthrough · editable scores
    Read the case study →Try the Relay demo ↗
§ 04Process

How the review works.

01

Agree on scope

We define the questions, available evidence and access needed for the review.

02

Review and test

I inspect the workflow and supporting code or data, test agreed cases and record what I find.

03

Review the findings

You receive the assessment and recommended next steps, including questions the review could not resolve.

§ 05Questions

Practical details.

What do I receive?
A written assessment with supporting examples, prioritized recommendations and any testing limits. The broader engagement also includes an options comparison and proposed roadmap.
What experience can I review?
The linked projects include source code, demos and implementation notes. They are independent work; test results and pending work are identified in each case study.
Do we need to meet?
A Reality Check is mostly asynchronous. A Full Diagnosis includes stakeholder interviews. We agree on meetings and review time when setting the scope.
What if the issue is in the workflow or data?
I include those findings in the assessment and explain how they affect the feature. Recommendations can include data or process changes.
How do we handle access and data?
We agree on access, permitted tools and data handling before work begins. Read-only access is the starting point; a sample or redacted dataset may be enough for the review.
Start here

Describe what you need reviewed.

Send a brief description of the feature or decision, the problem you are seeing and your timeline.

Start on Contrakristen@kristenmartino.ai