Concierge forward deployed engineering

AI-native software engineering in healthcare.

Senior forward deployed engineers, embedded in the codebases that run healthcare. We ship production software, AI agents, and EHR integrations, from blank repos to twenty-year monoliths. Startup pace, enterprise rigor.

Shipping stackagent-native · v1
Mode
Scale
01
Repo scaffold
monorepo · CLAUDE.md · skills
02
CI/CD
Actions · preview envs · trunk
03
Infra + IaC
Vercel · Supabase · Pulumi
04
Agent harness
MCP · Claude Code · tool registry
05
Observability
OTel · evals · runbooks

Blank repo, day one. We ship the scaffolding agents need to work, and the rigor your auditors expect.

US · EU · Asia · follow-the-sun

Shipping stack variants

Greenfield, startup

Blank repo, day one. We ship the scaffolding agents need to work, and the rigor your auditors expect.

  • 01 Repo scaffold: monorepo · CLAUDE.md · skills
  • 02 CI/CD: Actions · preview envs · trunk
  • 03 Infra + IaC: Vercel · Supabase · Pulumi
  • 04 Agent harness: MCP · Claude Code · tool registry
  • 05 Observability: OTel · evals · runbooks

Greenfield, enterprise

Blank repo, day one. We ship the scaffolding agents need to work, and the rigor your auditors expect.

  • 01 Repo scaffold: polyrepo · ADRs · golden paths
  • 02 CI/CD: gated promo · SLSA · SBOM
  • 03 Infra + IaC: AWS · Terraform · VPC · BAA
  • 04 Agent harness: MCP · scoped tools · audit log
  • 05 Observability: OTel · SIEM · 24/7 on-call

Brownfield, startup

Twenty-year monolith, day one. We lift it onto agent-native rails without a replatform.

  • 01 Repo scaffold: agent overlay · context maps
  • 02 CI/CD: parity tests · canaries · revert
  • 03 Infra + IaC: lift to IaC · ephemeral stages
  • 04 Agent harness: MCP wrappers · legacy shims
  • 05 Observability: bridge traces · regression evals

Brownfield, enterprise

Twenty-year monolith, day one. We lift it onto agent-native rails without a replatform.

  • 01 Repo scaffold: agent overlay · service maps · ADRs
  • 02 CI/CD: shadow runs · canary cohorts
  • 03 Infra + IaC: drift detect · golden images · BAA
  • 04 Agent harness: MCP · policy gates · PHI scopes
  • 05 Observability: OTel · SIEM · merge-gate evals
Built with
Mark Cuban Cost Plus DrugsTruepill
SteadyMD
Johnson & Johnson
Optum
Kenvue
Lithos AI
Stanford University
What we do

Six things,
done seriously.

Widal exists because most healthcare software teams don't need another vendor. They need senior engineers who understand FHIR, agentic systems, and the failure modes of clinical workflows, embedded alongside their team or running the build standalone, until the work is in production and measured.

The foundation

One stack underneath all of it.

Same architecture under every engagement. Healthcare data on the ingress side. A Model Context Protocol layer that keeps agents inside policy. A pod that plans, acts, and verifies. An eval harness that holds the line on the way back into your EHR.

Ingress01FHIR R4Epic · Cerner · athena · +1802HL7v2Mirth · interface engines03Bulk + X12Claims · eligibility04 · LayerMCP ServerTools · ResourcesPolicy · Scopes05 · PodAgentPlan · Act · VerifyTool calls · Memory06 · GateEval HarnessRegression · GuardrailsSchema · RefusalsTool use ↺Egress07 · ExitEHR EgressSMART · Audit · BAAStatus · ProductionWidal · Technical Foundation · v1Obs · Traces · Metrics · OTel
The playbook

Agent-first by default.

Every repo we drop into gets the same scaffolding. A versioned context file at the root. Slash commands and skills the agent can actually use. An eval suite that gates the merge. The diagram above isn't a sales pitch, it's what the directory tree looks like.

Playbookagent-first repo · examples
01 / 07

AGENTS.md

Project context

A version-controlled context file at the repo root that every coding agent reads first. Project map, conventions, the commands you actually run.

# Project: Acme Intake

## Stack
Next.js 15 · Postgres 16 · FHIR R4 · MCP server · Temporal

## What this repo does
Intake forms write evidence to typed `EvidenceFact` rows.
A deterministic policy engine reads them, never the LLM.
Triage decisions are authorized by `policy/engine.ts`.

## Commands
pnpm dev          → local stack
pnpm test:evals   → adversarial eval suite (must pass to merge)
pnpm policy:diff  → compare policy bundles

## Invariants
- The LLM never authorizes a workflow action.
- User text never updates `policy/*` files.
- Every PR runs the eval suite before merge.
Pause: hover · Tabs: clickFurther reading · Stanford CS146S · Anthropic Claude Code docs
  • FHIR R4 · Bulk FHIR · SMART-on-FHIR
  • HL7v2 · CDA · X12
  • Model Context Protocol
  • Eval harnesses + tracing
  • HIPAA + SOC 2 Type II controls
  • BAA included
Read the technical foundation
Agents

Agentic frameworks, in production

Tool use, evals, and observability. We've shipped multi-agent systems against real EHR data, not toy demos.

MCP

Model Context Protocol servers

Custom MCP servers wired into clinical systems so agents can read the chart, place orders, and stay inside policy.

FHIR

FHIR R4 + HL7v2 interop

Direct against Epic, Oracle Cerner, Athenahealth, eClinicalWorks, NextGen, AdvancedMD, Healthie, and the rest of the long tail. Bulk FHIR, SMART-on-FHIR, and the legacy HL7v2 work nobody wants to do.

Security

HIPAA + SOC 2 by default

BAA included, PHI scoped, row-level audit, your compliance envelope. Privacy-preserving inference where it matters.

Frontier labs we
work against
  • 01AnthropicClaude
  • 02OpenAIGPT
  • 03GoogleGemini
EHRs we work
against

21 systems, from the consolidators to the practice-shaped vendors. FHIR R4 where we can, HL7v2 and screen-scrape where we have to.

  • 01Epic
  • 02Oracle Cerner
  • 03Athenahealth
  • 04eClinicalWorks
  • 05NextGen
  • 06MEDITECH Expanse
  • 07AdvancedMD
  • 08ModMed
  • 09DrChrono
  • 10Elation Health
  • 11Healthie
  • 12Canvas Medical
  • 13Akute Health
  • 14Hint Health
  • 15Cerbo
  • 16Tebra
  • 17Medplum
  • 18Oystehr
  • 19SimplePractice
  • 20Ultralight
  • 21Jane App

Sourced from Widal · coding-agent readiness

The engagement

A pod, scoped to a
workflow that matters.

Six to twelve weeks. Senior engineering throughout. A working system at the end, not a roadmap and not a pilot. You can renew, extend, or close the engagement cleanly when the work is done.

01Week 0

Architecture read

Two senior engineers, one week, a written read on what to build and what to retire. No team commitment yet.

02Week 1 to 6

Embedded build

A small pod drops into your codebase. They write production code, attend your standups, and own the workflow end-to-end.

03Week 6 to 12

Hand-off + own it

Whatever we built, you keep. Runbooks, evals, on-call shape. We stay on retainer if you want, not because you have to.

How we work

Three commitments
we don't bend on.

Senior

Senior engineering, all the way through

No bench staffing, no juniors shadowing. The person on the kickoff call is the person writing the merge commit.

Owned

You own what we ship

Source, infra, evals, hand-off docs. No vendor lock, no perpetual license, no platform tax baked into the bill.

Measured

Measured the way you measure

We pick the metric the operator already runs against: encounter cycle time, claims yield, alert precision. Then we move it.

“A pod is scoped to a workflow, not to a vendor deliverable. We build what the operator actually needs, and we measure it the way they measure it.”

Open a thread
Open a thread

Tell us about the
workflow.

The first conversation is two senior engineers on a 30-minute call. No SDRs, no discovery deck. We read what you send, and we come prepared.

Response
Within one business day, written by an engineer.
Coverage
HQ in San Francisco. Engineers across the Americas, Europe, and Asia. Three regions, follow-the-sun, embedded with you.
0 / 2000

By sending, you agree we may reply by email. We don't add you to a marketing list.