10xAnd
Scoping review

Tools

We build the tooling. Some of it is free.

Nineteen free tools on public records, running now, and the software that runs inside the paid lines. The service-line demonstrations use synthetic data against a fictional device — never a client's file.

Nineteen free tools are live. They run on public records in your browser, with no account and no charge, and each carries a badge in its own chrome saying what it reads. The Opportunity Analyzer scores against a rubric whose wording is still a marked draft — it says so on every view. The service-line captures further down are previews, not live software, and are badged as such.

Free tools

Public data, no account, no charge. Handing you something that works is a better argument than a page about how rigorous we are.

Regulatory & FDA data

Design controls & quality

Engineering & concept

Labeling, UDI & interoperability

  • UDIScope

    Medical device barcode reverse-decoding, GS1 check-digit verification, and live FDA Global UDI Database (GUDID) intelligence.

    Open the tool →

  • MedLabel Studio

    Medical device labeling, GS1/GTIN-14 generation, ISO 15223-1 symbol placement, and 21 CFR Part 801 compliance linting.

    Open the tool →

  • FHIRBridge MD

    Convert medical device barcodes (UDI/GUDID) into compliant FHIR R4 resources, simulate SMART on FHIR app launches, and inspect EHR telemetry bridges.

    Open the tool →

  • EHR Integration Studio

    Interactive architecture designer, US Core FHIR payload builder, and Model Context Protocol (MCP) AI agent connector for Epic, Cerner, and TEFCA.

    Open the tool →

More

  • Prototype Atlas

    Describe or upload a part. Get a live budgetary quote, material options and a ranked shortlist of prototype shops near you and abroad.

    Open the tool →

Sourcing

Architecture & Standards

EHR Integration for Medical Devices & AI Agents

The pragmatic engineering realities of connecting medical devices, SaMD, and autonomous AI agents to Epic, Oracle Cerner, and national networks: SMART on FHIR v2 handshakes, US Core profiles, continuous waveform telemetry via SampledData, TEFCA networks, and Model Context Protocol (MCP) tool mediation.

Read the engineering guide →

Inside the service lines

These run inside paid engagements. They are shown for one reason: "AI-assisted with a named human reviewer" is a claim, and a claim about how software behaves should be watchable. Every one of these captures ends on the same beat — the export stays blocked until a person has signed for it.

GapScan MD

Design preview

Scores every design-control requirement against the clause it comes from, cites the evidence down to document, section and line, and refuses to export until a named reviewer has dispositioned each finding.

A rendered preview of the intended interface. Not running software.

A preview of GapScan MD. A readiness ring shows 76 over a synthetic design history file for a fictional device, with 27 requirements closed, 15 open findings and 3 needing human review. A coverage grid scores 42 requirements across 7 families as pass, partial, missing or cannot assess. One finding opens to show its evidence citation down to document, section and line. The review gate then reads EXPORT BLOCKED with 15 findings unreviewed, a reviewer types their name, dispositions climb to 15 of 15, and the gate turns green reading ALL FINDINGS DISPOSITIONED, REVIEW RECORD COMPLETE, at which point the export button unlocks.

Watch it run 1:12

Working tool

Recorded from a real run of the engine over a synthetic design history file for a device that does not exist. The chrome says so on every frame.

No audio track · every line is captioned on screen · nothing is fetched until you press play

Transcript (14 lines)
  1. 0:00 Before an audit, someone reads the whole file.
  2. 0:04 Usually weeks of reading. One pass.
  3. 0:08 How ready is this file, right now?
  4. 0:12 Synthetic DHF. 15 documents. Fictional device.
  5. 0:18 The file that holds up audits and submissions.
  6. 0:21 One command reads all 15 and sorts them by type.
  7. 0:26 Checked against QMSR design controls. Public sources.
  8. 0:32 12 requirements scored, with confidence and effort.
  9. 0:37 Every finding cites document, section, and line.
  10. 0:43 No evidence, no score. It just says so.
  11. 0:47 Safety-critical items are never scored by the tool.
  12. 0:52 Export attempted with findings unreviewed.
  13. 0:58 A named reviewer signs off on every finding.
  14. 1:04 DHF Gap Scan. Fixed price in 24h, delivered in 72h.

Edited before publishing: Final line trimmed from the recording.

TOOL-01 Synthetic design history file, fictional device. Two moments carry the whole argument: a requirement with no evidence is scored CANNOT ASSESS rather than guessed, and the export stays blocked until a named human has dispositioned every finding.

What must stay true of this tool, or the demonstration comes down:

  • A requirement with no evidence is scored CANNOT ASSESS, never guessed.
  • Every finding cites its evidence to document, section and line.
  • Safety-critical items are never auto-dispositioned by a bulk action.
  • Export is blocked until a named reviewer dispositions every finding.

It runs inside Remediate 10x.

FMEA Forge

Design preview

Drafts the risk file from a curated harm library rather than from free generation, and leaves severity blank wherever a human has to set it.

A rendered preview of the intended interface. Not running software.

A preview of FMEA Forge. A readiness ring shows 31 for a synthetic device risk file: 8 failure modes drafted, 3 severities withheld, 0 signed. The device decomposes into functions and components, with harms drawn from a versioned library rather than generated. Eight rows of failure modes are drafted, and three severity cells are deliberately left blank with the note that a model may not score patient harm. A red-flagged item shows severity tied to patient harm being set by a named person. The review gate blocks export until every row carries a named disposition, then turns green and the risk file exports with its review record attached, handing TestForge its input.

Watch it run 1:11

A rendered preview over a synthetic device with fictional harms and fictional reviewers.

No audio track · every line is captioned on screen · nothing is fetched until you press play

Transcript (14 lines)
  1. 0:00 Risk files are the gap we find most often.
  2. 0:04 Watch one drafted in hours, then signed line by line.
  3. 0:09 One number: how much of the risk file is finished?
  4. 0:14 This one just failed its risk section.
  5. 0:20 Harms come from a curated library. Nothing is invented.
  6. 0:26 Eight rows, each scored for occurrence and detection.
  7. 0:32 Three severity cells are left blank on purpose.
  8. 0:35 Severity tied to patient harm? The tool won't score it.
  9. 0:43 A person types that number. Their name goes on the row.
  10. 0:47 Nothing exports until every row has a reviewer's name.
  11. 0:51 Every row signed off. The light goes green.
  12. 0:55 Worksheets, a risk memo, and the file TestForge reads next.
  13. 1:00 Risk file drafting. Fixed scope, quoted in 24 hours.
  14. 1:05 Drafted by machine. Signed by a person.

TOOL-02 Synthetic device, fictional harms and reviewers. The blank severity cells are the point: the model drafts the row, and a person sets any severity tied to patient harm.

What must stay true of this tool, or the demonstration comes down:

  • Harms come from a curated, versioned library — never free generation.
  • Severity tied to patient harm is never auto-scored; a person sets it.
  • A red-flagged item is never dispositioned by a bulk action.
  • Nothing exports until every row carries a named disposition.

It runs inside Develop 10x.

TestForge

Design preview

Closes the verification loop: plan, protocols, fixture data and report as one chain, with a human review gate before anything ships.

A rendered preview of the intended interface. Not running software.

A preview of TestForge. It frames verification as four jobs — two writing, two execution — then shows a closed loop of eight nodes and one gate: design inputs, verification plan, draft protocols, automated fixture, results file, draft report, human review, signed package. A verification plan matrix lists requirements with test method, sample size, size rationale and acceptance criteria, marked planned values with nothing measured. The review gate blocks the signed package until a named reviewer dispositions every item, then turns green and the loop closes with the review record inside the package.

Watch it run 1:08

A rendered preview. The video says it out loud in its second line: nothing here is a working system yet.

No audio track · every line is captioned on screen · nothing is fetched until you press play

Transcript (12 lines)
  1. 0:00 Verification is four jobs. Two are writing. Two are execution.
  2. 0:08 Nothing here is a working system yet.
  3. 0:14 Four jobs, one loop, and one step a machine never takes.
  4. 0:19 The writing collapses first: the plan, then the protocols.
  5. 0:25 Method, sample size, why that size, acceptance criteria.
  6. 0:33 Automation runs the units. The rig writes the data.
  7. 0:40 It lands in the draft report untouched.
  8. 0:47 The loop stops here, on purpose.
  9. 0:50 A red-flag list marks what is never decided alone.
  10. 0:56 No named reviewer, no signed package.
  11. 0:59 Someone signs every item. The gate turns green.
  12. 1:03 The loop closes. The review record ships in the package.

Edited before publishing: Final line trimmed from the recording.

TOOL-03 Concept preview — nothing shown here is a working system yet, and the capture says so in its own chrome. The matrix carries planned values, not measurements.

What must stay true of this tool, or the demonstration comes down:

  • The loop does not close without a named reviewer.
  • Planned values are labelled as planned; nothing is presented as measured.
  • No protocol and no report leaves the loop without a name on the record.

It runs inside V&V 10x.

Pulse Engine

In development

The newsletter pipeline: drafts overnight from monitored sources, and sends nothing without a named human approval on the record.

Partly built. This capture is a preview of the intended interface, not a recording of a finished tool.

A preview of Pulse Engine, the internal newsletter tool. It shows a modelled founder-time budget of 37 minutes per week against a design target under 90 minutes, explicitly labelled modelled and not measured, broken into triage, editing and approval. A send queue shows three issues with three of three named approvals on record, each with an approval time, editor count and source count in the review record.

Watch it run 1:10

A rendered preview of intended UI, over sample content and fictional sources.

No audio track · every line is captioned on screen · nothing is fetched until you press play

Transcript (14 lines)

The 90-minute figure is a modelled design target. The screen it is spoken over labels it MODELLED, DESIGN TARGET and NOT MEASURED. It is not a measured result, and no week of work has been timed to produce it.

  1. 0:00 A weekly industry brief, drafted overnight.
  2. 0:03 Nothing sends without a name on it.
  3. 0:09 Monday, six a.m. Three briefs are written.
  4. 0:13 The judgment didn't.
  5. 0:20 The editor opens what the machine left behind.
  6. 0:24 Every claim traces back to a cited source.
  7. 0:28 Here the draft overclaimed.
  8. 0:32 The editor cuts it, and the cut stays on the record.
  9. 0:40 A named approval, or it does not send.
  10. 0:45 No name on a brief, no send.
  11. 0:49 Every brief approved by a person. The light goes green.
  12. 0:53 Under 90 minutes of founder time a week.
  13. 1:00 The automation does the hours. A human does the judgment.
  14. 1:04 This pipeline is also what we build for clients.

TOOL-04 Internal tool, sample content, fictional sources. The time figures are a modelled design target and are labelled as such — they are not a measured result.

What must stay true of this tool, or the demonstration comes down:

  • Nothing sends without a named human approval on the record.
  • The founder-time figures are a modelled design target, never presented as measured.

It runs inside Pulse.

Publishing the source

Some of this tooling will be released for anyone to run. That is a legal question before it is an engineering one — licensing, liability, and what a tool that touches a regulated file is allowed to say about itself — and it is not resolved. When it is, it lands here.