47HQ

Services

Three pillars.
Nine engagements.

Strategy, implementation, and infrastructure for funded teams shipping AI. Every engagement is fixed-scope with named deliverables and measurable outcomes.

Pillar 01

Strategic Guidance

Navigate AI complexity with fractional engineering leadership. Roadmaps optimized for ROI and a path to scale.

Pillar 02

AI Implementation

From copilots to fine-tuned models — production builds your engineering team owns.

Pillar 03

AI Infrastructure

Production-grade cloud infrastructure for AI/ML — optimized for cost, control, and scale.

Live demo reel

What we actually ship.

Every engagement leaves behind something you can open: a copilot, a retrieval index, a CI pipeline, a dashboard. Hover a tab to preview the artifact — click for the full engagement.

View AI Copilots engagement
support-copilot · prod
LIVE
JD
Why was my invoice charged twice last Tuesday?
AI
A retry hit our processor at 14:02 UTC after a timeout. The duplicate was auto-refunded within 4 minutes — no action needed.
billing.md#retriesincident-2148ledger.tx#9f3a
grounded · 3 sources412ms$0.0008
Hallucinations
0.31%
↓ 87%
P95 latency
412ms
↓ 38%
Citation rate
99.2%
↑ 12pt
Grounded answers, with citations.
47hq

How we work

Five steps. No surprises.

Every engagement runs the same playbook — so you know exactly what happens before you sign, and exactly what "done" looks like.

  1. 01

    Discovery

    Free 20-minute intro call. We get specific about the problem, your success metrics, and your timeline — and tell you on the spot if we're not the right fit.

  2. 02

    Diagnostic

    Paid 60–90 minute working session against your live system. 12-point assessment delivered as a written 1-page diagnostic plus session recording within 48 hours.

  3. 03

    Build

    Fixed-scope Statement of Work with named deliverables and a single price. Senior engineers ship in your repo, your cloud, against your eval harness. Weekly demos, not status decks.

  4. 04

    Ship

    Production handoff: runbooks, golden eval sets, on-call rotation guide, and the rollback paths your team will actually use at 2am. Code lives in your repos from day one.

  5. 05

    Metrics

    Every engagement ships against 3–5 measurable outcomes — MRR, p95 latency, eval pass rate, cost-per-query — agreed in writing before we start. We don't bill the last invoice until they're hit.

Pillar 01

Strategic Guidance

Navigate AI complexity with fractional engineering leadership. Roadmaps optimized for ROI and a path to scale.

01 · 2–4 weeks

AI Strategy

Fractional CTO expertise and roadmaps optimized for ROI and scale.

View full details
You're a fit if
  • Founder or CTO building an AI thesis with real budget behind it
  • Need an outside read on build-vs-buy and sequencing
  • Will translate findings into board-level decisions
Deliverables
  • AI roadmap with phased delivery and budget bands
  • Build-vs-buy analysis per surface
  • Hiring profile for the first 2 AI engineers
  • Risk register with mitigation plans
Outcome

A defensible 12-month AI roadmap your CEO and board can fund — with sequencing that respects what your team can actually ship.

Pillar 02

AI Implementation

From copilots to fine-tuned models — production builds your engineering team owns.

02 · 6–10 weeks

AI Copilots

In-product assistants grounded in customer data, with streaming, citations, and refusal logic.

View full details
You're a fit if
  • Live SaaS product with rich customer data
  • Need an in-product chat or assistant surface
  • Team that owns the codebase after handoff
Deliverables
  • Streaming chat surface with tool-use orchestration
  • Citation-grounded answer schemas
  • Per-tenant retrieval scoping and isolation
  • Refusal precision tuning + golden eval set
  • Cost-per-conversation budgeting and dashboards
Outcome

A copilot your customers actually use — with measurable wins on grounding, citation integrity, and cost per conversation.

03 · 4–8 weeks

Document Processing

Process thousands of structured and unstructured documents — complex layouts, tables, and forms.

View full details
You're a fit if
  • High-volume document workflows blocking ops or compliance
  • Mixed inputs: PDFs, scans, semi-structured forms
  • A reviewer-in-the-loop step you want to keep auditable
Deliverables
  • Document classifier and extraction pipeline
  • Reviewer queue with confidence-based routing
  • Schema validators and human-handoff hooks
  • Throughput + accuracy dashboards
Outcome

An auditable extraction pipeline with named accuracy on your document mix — and a reviewer surface your ops team trusts.

04 · 3–5 weeks

RAG & Embedding

Chunking, hybrid retrieval, reranking, and citation systems — measured against your golden set before they ship.

View full details
You're a fit if
  • Live RAG system shipped to paying customers
  • Willingness to instrument production with our eval harness
  • Single decision-maker on your side
Deliverables
  • Chunking redesign with three strategies benchmarked
  • Hybrid BM25 + dense retrieval with reranker
  • Citation system with source-grounding checks
  • Golden eval dataset of 500+ queries with LLM-as-judge scoring
  • CI gates so regressions block deploys
Outcome

Named before/after metrics on hallucination rate, retrieval precision @5, P95 latency, and cost-per-query — methodology your team can re-run forever.

05 · 4–8 weeks

Fine-Tuning & Inference

Specialized models that reduce cost and improve accuracy for your domain.

View full details
You're a fit if
  • A bounded task where general models cost or underperform
  • Access to representative labeled data (or a path to it)
  • Appetite for an eval harness to measure regressions
Deliverables
  • Dataset curation and labeling rubric
  • Fine-tune across two base models, benchmarked
  • Inference deployment with autoscaling
  • Eval suite + drift monitoring
Outcome

A specialised model that beats your current general-purpose baseline on accuracy and unit cost — with a path to keep it that way.

06 · 6–12 weeks

AI Agents

Multi-step agents that call your tools, your APIs, and your data — with deterministic rollback paths.

View full details
You're a fit if
  • Workflow with clear tools, APIs, and side effects to orchestrate
  • Engineering org ready for trace + replay infra
  • Guardrails treated as first-class, not an afterthought
Deliverables
  • Tool + API orchestration with typed schemas
  • Step-level trace + replay infrastructure
  • Per-step eval coverage, not just end-to-end
  • Rollback paths for every side-effect tool
  • Handoff package: runbooks, eval suite, rotation guide
Outcome

An agent that finishes the job — with traces you can replay, guardrails that fire when they should, and rollback when they don't.

Pillar 03

AI Infrastructure

Production-grade cloud infrastructure for AI/ML — optimized for cost, control, and scale.

07 · 4–8 weeks

Cloud Migration

Migrate from third-party model APIs to your own cloud — cut costs while gaining control and flexibility.

View full details
You're a fit if
  • Spend on third-party model APIs is material and growing
  • Need for data residency or VPC isolation
  • Engineering org that can own the resulting stack
Deliverables
  • Cost + capability comparison across providers
  • Cutover plan with shadow-traffic validation
  • Inference service deployed in your account
  • Unit-economics dashboard tracking pre/post cost
Outcome

Inference running in your cloud, under your control, with a named monthly cost reduction and a documented rollback path.

08 · 3–6 weeks

DevOps & Infrastructure

Auto-scaling, production-ready cloud infrastructure tuned for AI/ML workloads.

View full details
You're a fit if
  • AI workloads outgrowing a hand-rolled deployment
  • Need CI/CD that respects prompts, retrieval, and models as artifacts
  • On-call rotation that wants better signal
Deliverables
  • IaC for inference + retrieval services in your cloud
  • CI/CD with prompt + retrieval versioning
  • Autoscaling tuned to your latency and cost targets
  • Secrets, network, and access posture review
Outcome

Production infra your team can extend, scale, and on-call against — with deploy times measured in minutes, not days.

09 · 3–4 weeks

Production Telemetry

P95 latency, cost-per-query, drift, and eval regressions — on dashboards your on-call actually opens.

View full details
You're a fit if
  • AI feature already in production with real traffic
  • On-call rotation that needs better signal
  • Existing observability stack we can extend
Deliverables
  • Latency + cost dashboards per tenant
  • Embedding drift detection + alerting
  • Eval regression alerts on every deploy
  • Incident playbooks + rollback paths
  • On-call runbook tailored to your stack
Outcome

When your AI gets worse, your team knows in minutes — not quarters. Signal your on-call rotation actually trusts.

FAQ

Real questions, technically answered.

How fast can you start?
Most engagements kick off within 1–2 weeks of a signed Statement of Work. The discovery call is usually the same week you ask.
Do you work fixed-scope or time-and-materials?
Default is fixed scope with named deliverables and a fixed timeline. We will do retainer or T&M for incident-response or ongoing reliability work, never for greenfield builds.
Can we use our own LLM provider and vector DB?
Yes. We are model- and infrastructure-agnostic. Shipped against OpenAI, Anthropic, Bedrock, Vertex, self-hosted Llama, and against Pinecone, Weaviate, pgvector, Turbopuffer, and Mongo Atlas Vector.
What happens after delivery — retainers, handoff, training?
Engagements end with a written handoff package: runbooks, eval suite, on-call rotation guide. Optional monthly retainer for incident review and roadmap input. No multi-year MSAs.
Do you sign NDAs and BAAs?
Yes to both. NDAs before any production data changes hands. BAAs for HIPAA-scoped engagements.

Next step

Not sure which engagement fits?

Book a discovery call. We'll map your situation against the pillars above and tell you — honestly — whether we're the right partner.

Refundable if we're not a fitWritten diagnostic in 48 hoursSession run by a founder, not a sales rep