Assess. Diagnose. Treat.
Structured triage
for broken AI systems.
Your chatbot is drifting. Your OpenAI bill is doing weird things. A European customer just asked about AI Act compliance and nobody on your team knows what to say. Three symptoms, one real bug, no way to tell which. TriagAI runs a structured Health Check in 3–5 days, sorts the findings by severity, and hands you a plain-English report with effort and cost estimates for every recommended fix.
The problem nobody names
85% of ML projects fail to reach production. Of the ones that do, fewer than 40% sustain business value past 12 months. When your AI feature starts drifting, burning money, or attracting compliance questions, the options are bad:
- Big Four consultanciescost $50K+ and take three months for a 60-page PowerPoint.
- Upwork freelancersare a lottery.
- Observability platformslike Langfuse and LangSmith are great tools, but they are tools. They show you what is happening, not what is wrong.
What is missing is triage — the clinical discipline of rapid structured assessment under time pressure. Sort by severity. Name the real problem. Decide what can wait. Most AI failures are slow failures that never hit a pager, which is exactly when triage earns its keep. TriagAI is that triage.
How it works
Five steps from first call to final report. Total calendar time: one week. Total of your time: about two hours.
Discovery call
30 minutes, free. You describe the symptom. I decide whether this is an engagement I can take and quote a price on the call.
SOW and deposit
One-page SOW in your inbox within 4 hours of the call. You sign and pay a 50% deposit. Kickoff the next business day.
Diagnostic
3-5 business days. I run a structured diagnostic across five modules against your system with read-only API access.
Draft + walkthrough
Draft report delivered 24 hours before a 45-minute walkthrough call where we talk through findings and what to do next.
Final and follow-up
Final report, balance invoice, 7 days of email support, and a 30-day check-in to see how the fixes landed.
Pricing
Flat fees. 50% deposit at SOW signing, 50% on delivery. Invoices net 14. No retainers beyond Wellness. No per-seat anything.
Health Check
- ›Structured diagnostic across 5 modules
- ›Quality, cost, drift, integration, compliance
- ›PDF Health Check report
- ›45-minute walkthrough call
- ›7 days of email support
- ›30-day follow-up
Health Check + Compliance
- ›Everything in Health Check
- ›EU AI Act risk classification
- ›NYC LL144 and Colorado AI Act coverage
- ›Gap list mapped to regulatory articles
- ›Documentation inventory
- ›Prioritized remediation plan
Treatment Plan
- ›Hands-on implementation of specific fixes
- ›Scoped to a single finding or problem area
- ›Separate SOW, separate engagement
- ›Examples: prompt caching, model downgrade with eval, drift monitoring setup, EU AI Act doc build
- ›Never bundled with a Health Check
Wellness Retainer
- ›Quarterly Health Checks
- ›Drift monitoring with alerts
- ›Monthly AI spend review
- ›Same-day email response
- ›Three tiers: Baseline / Active / Embedded
- ›Cancel any time with 14 days notice
What you get in a Health Check
A 15-to-25-page PDF report. Plain English. No slideware. Structured so your engineering team can act on it the same day they read it.
Executive Summary
Health grade (A–F), top findings, compliance status, and a recommended-next-steps list your CTO can read in 2 minutes.
System Overview
Models in use, monthly spend, query volume, and the observable architecture of the AI feature under test.
Quality Assessment
Eval pass rates, notable failures, drift indicators, and a narrative on whether the system is getting better or worse.
Cost Analysis
Spend breakdown by model and endpoint plus optimization opportunities with dollar estimates for each fix.
Compliance Posture
EU AI Act risk classification, documentation gaps, NYC LL144 and Colorado AI Act coverage where applicable.
Prioritized Recommendations
Ranked by impact, each with effort estimates, cost estimates, and a risk rating. Everything actionable, nothing vague.
Why it works
It is a service, not a product.
Langfuse is a tool. TriagAI is a diagnostician. The value is the human judgment that translates raw telemetry into: here is what is wrong and here is what it costs to fix.
No platform to sell.
No SaaS. No subscription. No procurement conversation. Everything runs on one operator's laptop with read-only API credentials. An engagement starts with a 30-minute call and a one-page SOW.
Diagnostic, not prescriptive.
TriagAI does not sell a methodology. It runs five structured playbooks and reports findings. The client is free to fix things themselves, hire a Treatment Plan, or do nothing. The Health Check stands on its own.
Mid-market gap.
Too cheap for Accenture to care about, too structured for an Upwork freelancer to deliver reliably. The space between a $50 GitHub tool and a $50K audit has no occupant. That space is the entire market.
Limited offer
Launch partner engagements — 50% off
First ten clients receive a Health Check at $750 flat in exchange for a written testimonial and permission to publish an anonymized case study. Valid through May 15, 2026 or until all ten slots are filled.
Apply as a launch partnerFrequently asked
Are you an attorney?
No. Compliance findings are structured technical gap analyses, not legal determinations. You will still need qualified counsel.
What access do you need?
Read-only API keys for the providers in scope, ideally time-limited. I do not need production database access, shell access, or admin credentials.
How is this different from Langfuse or LangSmith?
Those are tools. TriagAI is a diagnostician. The tools show you a dashboard; I tell you what the dashboard means.
What if you break something?
The Health Check is read-only. Treatment Plans are scoped explicitly and nothing ships to prod without written approval. SOW caps liability at fees paid.
Do you work with enterprise clients?
No. Target is 20-500 person companies. I do not have the SOC 2 attestation or indemnification posture enterprise procurement typically requires.
Can I see a sample report?
Yes, on request after the discovery call. Not published publicly because reports contain pattern language the diagnostic relies on.
Get in touch
A 30-minute discovery call is free. I take one per day. You describe the symptom, I tell you whether it is an engagement I can take, and either send you a one-page SOW the same day or decline cleanly.
Not a good fit: Fortune 500 with a 40-page MSA, fractional MLOps staffing, penetration testing, or “AI” that is actually a rules engine. If any of those describe you, I will decline politely and point you at someone who does the thing you actually need.