Skip to content
Dunes

Services

Work on the systems under the models.

Four practices, one standard: change the architecture, the eval, the cost envelope, or the on-call path — or don’t bother. Starting fees below; a two-week diagnostic is $22,000.

01

Production AI systems

Most AI work fails after the demo. We design the system around the model: retrieval quality, tool use, evals, fallbacks, tracing, and the operational habits that catch drift before customers do.

  • Architecture for RAG, agents, and tool-using systems
  • Evaluation harnesses tied to product quality, not vibes
  • Observability, incident paths, and model-change playbooks
  • Cost and latency envelopes you can actually operate

02

AI infrastructure

We treat inference as a production service. Capacity, batching, failover, autoscaling, and spend are designed together so the platform does not surprise you the week traffic shows up.

  • Serving topology: batch, realtime, and mixed workloads
  • GPU / accelerator efficiency and cost attribution
  • Data paths for features, embeddings, and feedback
  • SLOs, capacity planning, and failure isolation

03

Architecture & platforms

Principal-level depth without a nine-month hiring cycle. We sit with your leads, write the hard documents, unblock the platform, and leave ownership behind — not a dependency.

  • Technical strategy that survives a board deck and a pager
  • Platform boundaries, paved paths, and team topology
  • Design reviews that change the system, not the slide
  • Hiring loops and leveling for AI / infra roles

04

Readiness reviews

A structured look at architecture, evals, cost, security posture, and operational ownership. You get a ranked risk list and a 30-day plan — not an 80-page PDF nobody reads.

  • Production-readiness score across system, ops, and org
  • Ranked risks with owners and effort
  • A 30-day plan with the one or two moves that matter
  • Optional follow-through as a build or retainer

Engagements

How the work is packaged.

  • Diagnostic

    2 weeks

    $22,000

    Fixed · 50% to start

    When the problem is real but the shape is not.

    Stakeholder interviews, architecture map, spend and failure review, written recommendation.

  • Architecture review

    1–2 weeks

    From $18,000

    Fixed · one named system

    A system exists. You need a sharp second opinion before the next build.

    Deep-dive review, decision record, sequenced changes, working session with leads.

  • Readiness audit

    2 weeks

    $35,000

    Fixed · go / no-go

    Launch, a large spend increase, or an agent going customer-facing.

    Checklist across quality, cost, security, ops. Ranked risks. 30-day plan.

  • Fractional

    Monthly retainer

    From $12,000 / mo

    3-month minimum · in advance

    Ongoing architecture, reviews, and platform leadership.

    Reserved hours, design reviews, hiring support, written decisions, Slack/on-call counsel.

  • Time-boxed build

    4–8 weeks

    From $55,000

    Fixed · 40 / 40 / 20

    A specific system needs to exist: eval harness, serving path, retrieval layer, control plane.

    Scoped delivery with your team. Code, runbooks, owners. No mysterious handoff.

  • Enablement

    1–3 days

    From $10,000

    Due on booking

    A team that needs a shared model of the work.

    Workshops on production AI, evals, inference cost, or platform design. Practical, not theatrical.

Start a conversation