Arviser
  • Open tools Promptfoo, DeepEval, Ragas, LangSmith
  • Fixed from-prices €2,400 PoC · €2,900 Readiness · €8,000 Audit
  • 10+ yrs Senior + AI Founders run the work with teams under our management
  • 50% Risk-Shared PoC Second 50% waived if narrow pipeline is unviable
  • Written-first scoping Free 48h form reply · Founder calls €250/h

Four clear directions

Start with one path. Prices and terms stay inside each direction where listed.

Examples and open research

Task, role, and an outcome you can open. Synthetic samples and open research — not invented client logos.

AI Security

Restricted-document exposure (synthetic)

Support assistant returns content from a file another team owns. Fix: enforce document permissions at retrieval, then re-test by role. Research context: WardBench.

Read sample finding · Request a security review

Build & deploy

Open proof and sprints

Founder benchmarks, PoC Go/Kill memos, and fixed-scope engineering sprints with published terms.

Open proof · 24h PoC · Deploy sprints

What a quality scorecard looks like

Synthetic example used on the home page and sample audit. Live audits use your thresholds.

Request a quality audit

Why pilots stall

  • Demo works. Real traffic and edge cases do not.
  • No shared quality numbers. Product and risk argue past each other.
  • EU AI Act and documentation questions arrive late.
  • Hiring strong AI people takes months.

Demo vs production · 24h PoC vs 40h Prototype · 50% Risk-Shared PoC protocol · Senior AI vs vibe-coding · Shared metrics · Hiring AI-QA · Independent scoring · Act evidence · Glossary

Industries

E-commerce

Typical risk: catalog hallucinations and weak refusal on abuse scripts.

Shared metrics

Health-tech

Typical risk: quality and docs without clinical claims or patient data on forms.

Scope and limits

Our technology stack

We pick tools that fit your product and stay maintainable for years. Full stack pages live under Engineering.

Eval tools we ship with

Open harnesses for golden sets, groundedness, refusal, and CI gates. Not a proprietary lock-in platform.

PfPromptfoo
DEDeepEval
RgRagas
LSLangSmith
Hugging Face

Cloud credit environments

Startup Program deploys inside your AWS, Google Cloud, or Azure credits when available.

AWS
Google Cloud
AzAzure
Docker

Open proof platforms

Named artifacts you can inspect on public hosts. Same founder. Commercial work via Arviser packages.

GitHub
Hugging Face
Zenodo
PyPI

Packages by direction

Details and from-prices inside Audit, Build, and Deploy — same four directions as the first screen (plus AI Security). Senior engineers with 10+ years production experience under founder oversight.

Audit · verify existing AI

from €2,900

AI Readiness Check · 1 week

Scorecard, risk register, metrics draft, kill or continue, 90-day roadmap.

Ideal for: stalled pilot, need gates before more spend.

Out of scope: ideation workshop only; full build; legal conformity opinion.

Details

AI Security: AI Security Review (scoped after intake) · narrower CI slice: LLM security lite

Build · build new AI

from €2,400

Technical PoC Research Spike · 24 hours

3–5 workdays. Verify in principle whether a working pipeline solves your core or narrow bottleneck across 20–50 cases and 2–3 models, with p95 latency and unit-cost math.

50% Risk-Shared Option: If the narrow pipeline is not feasible (Kill), the second 50% is waived (€1,200 total under our non-exclusive IP and sanitized research protocol). On Go: 100% exclusive IP + 100% credit to 80h MVP or Audit in 60 days.

Details · 50% Risk-Shared protocol

from €3,900

Working AI Prototype Sprint · 40 hours

1–2 weeks. Deployed end-to-end vertical slice (UI or API integration + LLM/RAG backend), basic guardrails, and 15–25 smoke evals in Promptfoo.

Ideal for: clear ideas where a separate PoC is not needed and you want a clean, working build without junior vibe-code debt.

Who builds: 10+ yr Senior Engineer with AI + Founder review. (€3,300 co-branded startup option).

Details · 24h vs 40h guide

Deploy · harden and staff

from €3,100

Hardening, CI Harness & Pilot-to-Production

Fixed-scope senior sprints or custom post-audit builds: RAG, agents, integrations, CI eval gates, and production monitoring.

Ideal for: taking a prototype or audited pilot into production.

Fixed sprint tiers (EUR, excl. VAT)
SprintHoursFrom
Eval & Guardrail CI Harness32h (1 wk)€3,100
Prototype-to-Production Hardening60h (2 wks)€5,600
Production AI MVP Sprint80–120h€8,500
Full Pilot-to-Production4–8 wksscoped

All AI sprints · Request scope

from €6,500/mo

Dedicated specialists, Retainers & Founder Advisory

AI/ML, AI-QA, MLOps full-time (min 3 months, EU time zone, replacement in 5–10 working days), Fractional Senior AI Architect retainers (20h/mo from €2,400), or Direct Founder Consultations (€250/h).

Ideal for: sustained capacity after a gate, sprint, or audit, or targeted executive architecture review.

Out of scope: unpaid discovery workshops; unpaid trials.

From-rates (EUR, excl. VAT)
RoleFrom
Direct Founder Consultation (100% credited in 30 days)€250 / hour
Senior AI Architect Retainer (20h / 40h)€2,400 / €4,500 / mo
Mid AI-QA€6,500 / mo
Mid AI/ML€7,500 / mo
Senior AI-QA and MLOps€9,000 / mo
Senior AI/ML (10+ yrs)€10,000 / mo

Request scope or session

Outputs: quality metrics, failure modes, working code, and repeatable suites for stated software behaviour. Not clinical evidence under MDR/IVDR. Not AI Act certification.

100% of the Readiness fee credits toward the next signed Arviser engagement within 90 days. LLM Quality Audit is billed on its own and is not credited forward. 100% of the 24h PoC fee credits toward an 80h MVP, Pilot-to-Production, or Audit within 60 days.

Audit and Sprints: overrun inside the written scope is on us. Without a prioritized roadmap and working eval baseline for the agreed surface, you do not pay the Audit fee.

On the 24h PoC Risk-Shared Track, if the narrow pipeline is not feasible across 2–3 tested models, the second 50% payment is waived. Payment is 50% on signature and 50% on delivery.

Founder

Vadim Nareyko, founder of Arviser

Vadim Nareyko

30+ years in IT. C-level path. AI/ML leadership. Warsaw. Founders run the work with senior engineering teams (10+ years experience and daily AI-native workflows) under our management.

  • Written scope and Q&A replies within 48 hours on workdays (free via form)
  • Direct 1-on-1 Founder Consultation: €250/hour (100% credited to any package within 30 days)
  • Arviser Sp. z o.o. · KRS 0001068182 · [email protected]

Open proof

Named capability work you can inspect. Not client case studies.

WardBench

Synthetic FHIR agent benchmark. Usefulness versus policy violations under role, purpose, and Consent.

Open research · Synthetic data · Not a compliance certificate

GitHub · DOI 10.5281/zenodo.23664058

Same founder. Commercial work via Arviser Audit and Readiness.

Genivara and Genivara-PII

On-device privacy and PII detection before data reaches the model. Demo and R&D track.

R&D demo · Before the model · Not a medical device

8.2 MB · ~15 ms WASM · on-device

Hugging Face · ONNX · Teacher · DOI 10.5281/zenodo.19118893

lexaudit

Linguistic audit CLI for calques, false friends, and invented words in LLM or translated prose. Detection-only. MIT tool by the same author (FairHelm monorepo).

Open tool · Not LeakAudit · Not LexAudit contracts

Used in FairHelm CI and planned for Arviser localized copy. Commercial delivery still goes through Eval CI and AI Sprints.

SeaLights (employer framing)

Quality intelligence and test-impact analytics experience: shipping under measured test risk and CI insight. Role and domain framing only.

Employer proof · No NDA client names · No invented metrics

Ties to Eval CI harness and AI code modernization characterization gates.

WardBench local v0.4 draft: utility versus policy violation rate across models on synthetic patients
WardBench local v0.4: utility versus violation rate (synthetic patients).
Genivara-PII architecture: distillation and pruning to an 8.2 MB ONNX model running in the browser WASM runtime
Genivara-PII Fig. 1: teacher to nano, WASM, Hybrid Consistency.

How we work

  1. Assess or Spike Days 0–5

    Written scope form and Q&A (free, 48h reply), then 24h PoC Spike, Readiness, or Audit. Artifact: written scope and pass marks.

  2. Measure or Prototype 1–2 weeks

    Evidence, scores, and roadmap, or a 40h working AI prototype with smoke evals. Artifact: scorecard or deployed repo.

  3. Harden, build, or staff Optional

    60h Hardening, 80h Production MVP, Pilot-to-Production, or embedded specialists. Artifact: scoped delivery plan.

  4. Gate Before scale

    Regression evals before scale. Artifact: CI gate or scheduled suite.

We sell the decision, the senior engineering build, and the evidence pack, not a proprietary lock-in platform.

Compared to alternatives

Arviser Big4 Red team Workshop
IndependenceYesPartialYesVaries
Duration24h to 2 weeksMonthsDays to weeksDays
Public from-priceYesRareRareOften
Eval evidenceScores + suiteBoard docsAbuse pathsSlides
Go or no-goExplicitVariesThinRare

EU AI Act dates we track

Orientation for counsel. Not legal advice. Not certification. As of October 2026.

FromWhat
2 Aug 2026Art. 50 transparency obligations apply
2 Dec 2026Machine-readable marking for generative systems already on the market

Full dates table · optional technical evidence add-on on the LLM Quality Audit.

Written scoping (free) & Direct Founder Consultation (€250/h)

To keep senior engineering focus on active deliverables, we do not run unvetted discovery calls. You have two clear paths:

  • 1. Written Scope & Technical Q&A (Free · 48h reply): Send your system context, package question, or constraints via the scope form below. A founder or 10+ year senior engineer replies in writing with a package fit check, fixed price band, and answers within 48 hours on workdays. When a package scope is ready to sign, we schedule a brief 15-minute alignment call.
  • 2. Direct 1-on-1 Founder Consultation (€250 / hour excl. VAT): Need live architectural review, RAG or agent troubleshooting, evaluation design, or executive AI strategy with a founder (30+ years in IT) without committing to a full package yet? Book a paid 60-minute working session. 100% of the €250 session fee credits toward any signed Arviser package (from €2,400) within 30 days.

Health-tech

Measurable LLM quality and production documentation for health-tech product companies.

FAQ

Readiness or Audit first?

Start with Readiness if you need gates and kill or continue in one week. Start with Audit if a live system needs scores and a repeatable suite. Comparison table.

24-hour PoC Spike or 40-hour Working Prototype?

Start with the 24-hour Technical PoC Spike (from €2,400) if you have a hard technical feasibility, accuracy, latency, or unit-cost risk. Start with the 40-hour Working Prototype Sprint (from €3,900) when feasibility is clear and you need a deployed vertical slice with baseline evals. PoC vs Prototype guide.

How do calls, questions, and the €250/h Founder Consultation work?

Submit any project scope or technical question via the form below and receive a written senior review and fixed price band within 48 hours for free (followed by a 15-minute alignment call when a package scope is ready).

If you want a live 60-minute architecture or strategy deep-dive with a founder before choosing a package, it is €250/hour excl. VAT, and 100% of that €250 fee credits toward any signed package (from €2,400) within 30 days.

How does the 50% Risk-Shared PoC discount work?

In Hours 1–4 we lock numeric pass marks for your core or narrow pipeline bottleneck across 20–50 cases and test 2–3 models. If a working pipeline is feasible in principle (Go), you pay the full €2,400, receive 100% exclusive IP under NDA, and can credit 100% toward an 80h MVP or Audit in 60 days.

If none of the tested setups meet the pass thresholds (Kill), the second 50% is waived (€1,200 total): you receive a non-exclusive license and full report while Arviser retains harness rights and may publish a sanitized, synthetic-data benchmark note. Full protocol.

What special terms are available for startups?

Pre-Seed to Series A teams get access to our Startup & Scale-up Program: 50% Risk-Shared PoC, 15% Co-Branded Open Proof discount on 40h Prototype (€3,300) or Audit (€6,800), a free 2-page VC Technical Due Diligence Memo, and deployment inside your AWS, Google, or Azure startup credits.

Hiring AI-QA takes months. What then?

Measure with Readiness or Audit while you hire, book a fixed-hour Senior AI Sprint, or book dedicated specialists from €6,500/mo. Hiring path.

EU AI Act?

Optional add-on: technical tests and documentation for your counsel. Not legal advice. Not certification. Dates we track · Marking window.

NDA and data?

NDA before sensitive material. Prefer your environment. No patient data via website forms.

Credit?

100% of the Readiness fee credits toward the next signed Arviser engagement within 90 days. LLM Quality Audit is billed on its own and is not credited forward. 100% of a confirmed 24h PoC fee (€2,400) credits toward an 80h MVP, Pilot-to-Production, or Audit within 60 days (or 50% toward a 40h Prototype).

100% of a 1-hour Founder Consultation (€250) credits toward any package within 30 days. One credit path. Then expires.

Tell us about your project

Free written scoping and technical Q&A reply within 48 hours on workdays (Warsaw), or select Direct Founder Consultation (€250/h) below. Controller: Arviser Sp. z o.o. Used to respond and take pre-contract steps. Privacy. No patient data.

1-click prompt templates:
Optional details (role, industry, timezone)

Discuss your project