SLED Subcontracting
SERVICE 02 · AGENTIC ENGINEERING

AI agents that survive contact with production

Retrieval, tool use, and voice agents, shipped with evaluations, approval gates, and a cost ceiling.

Let's talk See what we build
  • Evaluation gate
  • Human gate on irreversible actions
  • Cost ceiling

WHAT WE BUILT

Four things we build, one senior bench

Workflow agents, retrieval, voice, and the integrations that make them useful. Each ships with the proof layer that keeps it in production.

01

Workflow agents

Read context, call your tools, and complete multi-step processes with exceptions handled and escalated.

Proof layer tool-use loops with typed schemas, idempotent actions, retry and backoff policy, state stored separately from the execution surface.

02

Retrieval systems

Over your own documents and data, with citations back to the source document on every answer.

Proof layer chunking and embedding strategy chosen per corpus, hybrid retrieval where lexical matters, reranking, answer grounding checks.

03

Voice agents

For intake, qualification, and follow-up, with full transcript logging.

Proof layer telephony, barge-in and latency budgets, structured extraction into your CRM, escalation on defined triggers.

04

Integrations

Into your CRM, ticketing, ERP, and data stack through Model Context Protocol servers and native APIs.

Proof layer MCP where a governed tool surface is worth the overhead, native APIs where it is not. We do not add a protocol layer to justify a buzzword.

Named stack
Claude OpenAI LangChain MCP RAG Pinecone pgvector PostgreSQL n8n Zapier Make Retell ElevenLabs AWS GCP Python TypeScript Docker
from kickoff to one workflow live in production
6–8 weeks
graded evaluation set per workflow before launch
50–200 cases
automatic stop at the monthly cost ceiling
100% halt
of agent projects by 2027 (Gartner). We fix the 3 causes first.
40% canceled

Use cases

Work we can hand to an agent

If a person reads the same screen, queries the same data, and clicks the same tools all day, it is a candidate. These are the shapes we build most often.

  • Voice intake and scheduling

    Answers inbound calls, qualifies the request, books or reschedules, and writes a structured record into your CRM or EHR.

    Clinics · Home services · Legal intake

  • Support triage and reply drafting

    Reads each ticket or email, pulls live order or account context, drafts an on-brand reply, and escalates anything outside policy.

    E-commerce · SaaS · Property management

  • Document extraction to system of record

    Turns PDFs, forms, and scans into validated fields, then creates or updates the record with an audit trail on every write.

    Lending · Insurance · Logistics

  • Lead qualification and nurture

    Scores and enriches new leads, routes them, and runs personalized follow-up across email, SMS, and WhatsApp until a human takes over.

    Sales teams · Agencies · Real estate

  • Internal knowledge assistant

    Answers staff questions from your policies, contracts, and runbooks with a citation back to the source document on every answer.

    Operations · HR · Compliance

  • Order and exception handling

    Watches orders, shipments, or claims, resolves routine exceptions with your tools, and hands the rest to a person with full context.

    Retail · 3PL · Field service

  • Reporting and insight agents

    Assembles recurring reports from your data stack, explains the movement, and flags anomalies before the meeting, not after.

    Finance · Marketing · Executive teams

  • Review and verification

    Checks submissions, invoices, or records against rules and history, blocks what fails, and logs every decision as a regression case.

    Accounting · Procurement · Public sector

Client stories

Agents already doing the job

Each one started as a workflow someone worked by hand. Each one now runs with an evaluation gate, an escalation rule, and a cost ceiling.

All case studies

How we ship

Built to be evaluated, governed, and owned

Every agent leaves with the same four things defined in the specification before code exists.

01 Evaluation

A demo proves it can succeed once

Production requires knowing how often it succeeds, and on what.

The evaluation set
50 to 200 graded cases per workflow at launch, drawn from your actual history including the ugly ones.
Regression cases
Every caught failure becomes a permanent regression case. The evaluation set only grows, which is what makes an agent improve month over month instead of oscillating.
3 moments
Evaluations run before launch as a gate, on every change as a regression check, and on a schedule to catch drift when a model provider ships an update you did not ask for.
02 When it is wrong

Agents are wrong sometimes

The engineering question is what happens next.

Low confidence on a routine case
Escalates to a human with full context and the reasoning trace attached.
Retrieval finds nothing relevant
Says so and escalates. It does not answer from general knowledge.
Tool call fails or times out
Retries with backoff, then escalates. Actions are idempotent so retries cannot double-charge or double-send.
03 Governance

An engineering property, not a policy filed after launch

Every agent we build ships with these controls defined in the specification before code exists.

The audit trail
Every action is logged: what the agent read, which tool it called, what it produced, whether a human approved it, and when. Exportable without our help.
The approval matrix
Which actions run autonomously and which require a human, named by role. Changing it is a written change, never a silent config flip.
The data boundary map
What the agent may and may not read, and where each data class sits. A forwardable deliverable, not an internal note.
04 Run cost

The agent cannot spend more than you approved

And you get told before it gets close. Per-run token and infrastructure telemetry against a hard ceiling set in the specification.

Alert thresholds
70% and 90% of monthly budget, with an automatic halt at 100%.
Model tokens per run
Estimated against expected monthly volume, with the volume assumption stated so it can be challenged.
Retrieval and vector storage
Named in the estimate published before you sign.

WHO RUNS IT AFTER LAUNCH

Operate retainer We monitor, tune, run scheduled evaluations, and respond within an agreed window.
Handover Your team runs it. You get the runbook, the evaluation set, the dashboards, and 2 training sessions. We stay reachable for 30 days.

WHAT WE WILL NOT BUILD

  • Agents that take irreversible action on customers without a human gate at launch.
  • Agents over data we are not permitted to see.
  • Agents whose success cannot be written as a checkable sentence. If we cannot evaluate it, we will not ship it.
  • A single agent asked to do 5 unrelated jobs.
  • Anything where an existing tool solves it. We will tell you the tool.
For U.S. SLED prime contractors

AI agent and RAG capability, delivered as your subcontractor.

If your SLED scope calls for AI automation, RAG over a document corpus, or agent deployment, we build it behind the prime. The boundary is fixed on purpose.

NAICS 541511 541512 541519
See SLED Subcontracting

NDA-first, subcontract-only. We work behind the prime. We do not pursue prime contracts and we never face the agency.

Capability over claims. Model-agnostic agent architectures, retrieval systems, and workflow automation, mapped to your bid’s technical scope.

Governance built in. The audit trail, approval matrix, and data boundary map are part of every agent we ship, the controls a procurement security review asks for.

FAQ

The questions you were going to ask

Straight answers about agents in production. If yours isn't here, ask it on the call — we answer the hard ones first.

Ask the hard one

How do you stop it hallucinating in front of a customer?

Scoped tools, retrieval over approved sources only, automated verification on outputs, an evaluation gate before deploy, and a human gate on anything irreversible. When retrieval finds nothing, the agent says so instead of inventing an answer.

Which model do you use?

Whichever wins the evaluation on your cases. We are model-agnostic by contract and re-test as models change. A vendor that locks you to 1 model is optimising for their relationship, not your result.

Who owns the prompts, the evaluation set, and the code?

You do, all 3, in your repository. The evaluation set is often the most valuable artifact of the engagement and it is yours.

Is our data training someone’s model?

No. Enterprise API terms that exclude training, named in your specification.

How long until production, honestly?

6 to 8 weeks for 1 scoped workflow. If your case history is missing or your integrations are undocumented, we say so in scoping and quote the discovery separately rather than absorbing it into an optimistic timeline.

Can we show an auditor what the agent did?

Yes. Every action, tool call, output, and human approval is logged and exportable, and model, prompt, and evaluation-set versions are tracked so any output traces back to the configuration that produced it.

Related work

Service Data and platform → Service Custom software development → Service Forward deployed engineering → Industry Healthcare →

Start the conversation

Buildanagentthat
reaches production

Tell us the workflow someone works by hand all day. We will tell you whether an agent fits, and what it takes to ship it safely.

Scope an agent build
30 minutes the engineer who leads delivery no deck, no pitch