Arviser

Arviser

24-Hour Technical PoC Research Spike

Verify in principle whether a working AI pipeline can solve the core or narrow bottleneck of your idea in 24 senior engineering hours. Executed by software engineers with 10+ years of experience and AI-native workflows under founder oversight.

From €2,400 · 50% Risk-Shared Option (€1,200 on Kill) · 10+ yrs Senior Engineers + AI · Written scope in 48h · Founder call €250/h (credited)

When you need a 24-hour PoC instead of a prototype

Use this package when your idea has a real technical unknown: whether a model can parse your messy documents, hit p95 latency under 1.5 seconds, keep unit cost below your margin ceiling, or strip PII on-device before calling an external API.

Our goal is to verify in principle that a reliable pipeline can be built, even for a narrow and high-risk slice of the problem, before you spend on full product development.

The same measurement discipline shows up in our open engineering layers, including WardBench for agent safety after FHIR access. Commercial PoC gates stay separate from that research track.

If feasibility is already clear and you need a working UI or workflow slice, skip straight to the 40-Hour Working AI Prototype Sprint (from €3,900).

PoC spike vs prototype sprint vs agency discovery

24-Hour Technical PoC Spike vs 40-Hour Working AI Prototype Sprint vs Typical European Agency Discovery
Axis 24-Hour Technical PoC Spike (€2,400; €1,200 on Kill) 40-Hour Working AI Prototype Sprint (€3,900; €3,300 Co-Branded) Typical European Agency Discovery (€10,000 to €25,000)
Job to be done Prove or kill a narrow technical bottleneck on 20 to 50 cases Ship a deployed vertical slice with smoke evals Workshops, decks, and open-ended discovery
Who builds Senior engineers (10+ yrs) + founder review Senior engineers (10+ yrs) + founder review Mixed seniority; often PM-heavy
Timeline 24 hours across 3 to 5 business days 40 hours across 1 to 2 weeks Often 3 to 6 weeks before build starts
Published from-price €2,400 (~10,300 PLN); €1,200 on Kill €3,900 (~16,800 PLN); €3,300 Co-Branded €10,000 to €25,000 discovery before implementation
Primary output Go, Pivot, or Kill memo + benchmark harness Dockerized prototype + 15 to 25 Promptfoo cases Slideware and backlog; code optional

How the 24 hours are spent

  1. Hours 1–4 · Target slice, pass marks, and test data Day 1

    Isolate the core bottleneck or a narrow sub-task, agree in writing on numeric Pass thresholds (accuracy, groundedness, p95 latency, cost per 1,000 runs, privacy boundary), and prepare 20–50 representative or synthetic test cases.

  2. Hours 5–16 · Isolated testbed and multi-model comparison Days 2–3

    Build a clean, reproducible CLI or notebook harness. Test at least 2–3 distinct architectures or models (commercial APIs, open-weight models, hybrid retrieval, or local PII pre-filtering) without wasting hours on UI chrome.

  3. Hours 17–21 · Automated benchmark and edge-case stress test Day 4

    Run the candidate pipelines through Promptfoo or DeepEval against your 20–50 test cases, including adversarial prompts, missing-context refusals, and latency spikes.

  4. Hours 22–24 · Go, Pivot, or Kill memo and handoff Day 5

    Deliver the code repository, benchmark scoreboard, production unit-cost projection, optional VC Data Room Memo, and a 30-minute engineering review with the founder.

Why 10+ year senior engineers with AI make 24 hours enough

Traditional agencies quote 3–4 weeks for a PoC because junior developers spend most of the budget writing boilerplate and learning the stack. At Arviser, every spike is executed hands-on by senior software engineers with 10+ years in production backend, data, and distributed systems who use AI-accelerated coding and evaluation tools daily, reviewed by a founder with 30+ years in IT.

AI tooling handles the scaffolding in minutes. Your 24 hours go entirely into experimental design, retrieval and prompt engineering, failure-mode analysis, and repeatable benchmarks.

Two commercial tracks: Risk-Shared 50% Guarantee vs Closed NDA

Choose your preferred track in writing before kickoff:

Term Risk-Shared Open-Research Track Standard Closed-NDA Track
Kickoff deposit (50%) €1,200 €1,200
If narrow pipeline is feasible (Go) €1,200 balance · 100% exclusive IP under NDA €1,200 balance · 100% exclusive IP under NDA
If pipeline is not feasible (Kill) €0 balance (50% off · total €1,200) €1,200 balance (total €2,400)
IP & research on Kill Non-exclusive client license; sanitized synthetic benchmark rights for Arviser 100% exclusive IP transfer under closed NDA
Fee credit on Go (60 days) 100% (€2,400) to 80h MVP or Audit · 50% to 40h Prototype 100% (€2,400) to 80h MVP or Audit · 50% to 40h Prototype

How feasibility and sanitization work: Feasibility is confirmed (Go) when any of the 2–3 tested configurations achieves the agreed pass marks on the target pipeline slice.

On a Risk-Shared Kill verdict, you receive the full repository and 2–3 model benchmark logs with a non-exclusive license, while any Arviser research note uses 100% synthetic data, zero client identifiers, and a 48-hour client review window. Full protocol.

What you leave with

Reproducible PoC repository

Clean Python or TypeScript testbed with pinned configs, prompts, and scripts your team can run locally.

Multi-model comparison table (2–3 setups)

Side-by-side numbers across 2–3 tested approaches: task accuracy, groundedness, refusal rate, p95 latency, and token cost.

Promptfoo evaluation config (20–50 cases)

A working evaluation baseline that carries directly into your prototype or CI pipeline.

Go, Pivot, or Kill decision memo (+ optional VC Data Room Memo)

Written verdict with production unit-economics math and a concrete architecture recommendation for the next phase.

FAQ

Should we book the 24-hour PoC or the 40-hour Prototype?

Book the 24-hour PoC if you have an unproven technical bottleneck (complex document parsing, strict latency, tight unit cost, or on-device privacy). Book the 40-hour Prototype if you already know the LLM can do the task and need a deployed, clickable vertical slice. Full comparison.

What counts as a confirmed Go versus a Kill?

In Hours 1 to 4 we lock numeric pass thresholds for the core or narrow bottleneck of your pipeline across 20 to 50 cases. If at least one of the 2 to 3 tested model or retrieval setups meets those thresholds, confirming in principle that the pipeline can be built, the PoC is a Go.

If none meet the thresholds, it is a Kill and the second 50% fee is waived on the Risk-Shared Track.

What if our company wants full IP ownership even on a failed PoC?

You can select the Standard Closed-NDA Track at kickoff, or upgrade a Risk-Shared Kill to full exclusive IP at any time by paying the remaining €1,200 balance.

How do we handle sensitive or health-tech data?

We sign an NDA before kickoff and default to synthetic or anonymized datasets. If real data is required, we run inside your environment under a DPA. Never send patient data via website forms.

Request written scope or ask a question

Free written scoping and technical Q&A reply within 48 hours on workdays (Warsaw), or select Direct Founder Consultation (€250/h, 100% credited in 30 days) below. Controller: Arviser Sp. z o.o. Privacy. No patient data.

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

Discuss your project