AI audit reviewer
A draft financial-statement package goes in; a categorised exception report comes back in about 90 seconds, at a measured $0.24–0.28 per review.
- Python
- FastAPI
- Anthropic Claude
AI accuracy and cost audit 5–10 days
RAG and LLM cost audit: accuracy and cost per request
In 5–10 days I measure what your AI feature costs per request and how often it is right, then show which changes move both. On a CPA firm's review tool that measurement put a review at $0.24–0.28 and found a setting with more findings at 63% of the cost.
Who it is for
- Teams whose RAG answers or LLM output cannot be trusted yet.
- Products whose model bill is growing faster than their usage.
What the work covers
- Cost per request, measured on your real traffic or samples.
- Accuracy: where answers are ungrounded, and where code should check what the model claims.
- Failure handling: timeouts, refusals, rate limits and empty replies.
- A written report of findings, ranked by cost and risk.
Proof
Evoriqa
My own SaaS: a multi-tenant AI customer-support platform, live in production, built by the person who also pays its inference bill.
- Next.js
- TypeScript
- PostgreSQL
Questions
- What do you need from us?
- Access to the code, and real or representative inputs to run through it. Production credentials are not needed for the measurement.
- What does the audit produce?
- A measured cost per request, an accuracy picture, and the changes ranked by what they save or fix. The CPA tool's audit found a cheaper setting with more findings.
- Do you also make the changes?
- That is a separate piece of work if you want it. My preference is to agree it after the report, once the size of each change is known.
Tell me what you’re building and where it’s stuck.
I’ll tell you the cleanest path forward, including if it’s “don’t build that.”
Or write tocontact@alihassan.dev
