The Platform

Traceability as a database primitive, not a spreadsheet exercise.

Every requirement, BOM line, risk, test result, and supplier relationship is a node in one graph. An AI layer drafts and challenges; a human approval gate gates every change; a deterministic core holds the audit trail.

AI proposes. The graph enforces. The human decides.

On this page
Two layers, one gate

Deterministic core. Probabilistic intelligence. One explicit approval boundary.

Avera Core is the system of record — immutable, audited, 21 CFR Part 11 compliant. Avera AI sits above it and never writes directly. Every suggestion is queued for a human to approve, revise, or reject.

Probabilistic · Avera AI

Proposes, challenges, improves.

Drafts risk analyses, V&V protocols, submission sections. Claude via MCP. Context-scoped to the nodes it touches.

↓ Human approval gate ↓

Deterministic · Avera Core

System of record.

Immutable audit trail. Electronic signatures (eIDAS, ZertES). Graph-native traceability. 21 CFR Part 11.

Every AI suggestion requires human approval before entering the system of record.

No training on customer data. Every API call audited. Per-item opt-in at the node level.

The graph is the source of truth. The AI is an assistant with a leash. The human is the final approver of every change.

The moat

Graph-native, not relational.

Relationships are first-class citizens — not foreign keys. Impact analysis is a traversal, not a report. This is structural, not a feature.

Impact analysis

Change an input. See the blast radius in seconds.

Change a design input and every affected design output, V&V protocol, risk control, and submission section surfaces instantly. No JOIN queries. No nightly exports. No stale trace matrix.

MATCH (in:DesignInput {id: "DI-042"})-[:TRACES_TO*]->(x) RETURN x
Trace-current

The DHF is a live view, not a retroactive artifact.

Traceability is computed from the graph at read time. There is no trace matrix to reconstruct before an audit — the matrix is a projection of the graph, always current.

Avera AI

Proposes. Challenges. Improves.

Claude via MCP, scoped to the nodes it touches. Every output routes through the human approval gate. Every API call is logged.

Proposes

Drafts the volume work.

Risk analyses, V&V protocols, submission sections, test-procedure skeletons, FMEA entries. Generated from live graph context, not a blank page.

Challenges

Scans for orphans, gaps, and contradictions.

Continuously checks for orphaned items, missing traces, contradictions across DHF / DMR / technical files. A quality layer, not just a generator.

Guardrails

No training on your data. Per-item opt-in. Every call audited.

Customer data is never used to train any model. AI use is opt-in at the node level. Every transmission to Anthropic is logged with node scope, timestamp, and human reviewer.

LiveDocs

Documents that query the graph.

DHF, DMR, technical files, and submission sections authored as live documents — tables and sections bound to graph queries, updated as the underlying nodes change.

Bound to the graph

When the design changes, the submission knows.

510(k), eSTAR, and MDR technical files generated from live graph data. A queryTable inside a document pulls from the graph at render time, not at authoring time.

Windows Explorer mount

Any app, zero integration.

Avera appears as a mounted filesystem. SolidWorks, Altium, MATLAB, Jira, Confluence read and write through it — no plugin, no connector, no integration project.

Branching & variants

Git-style branches for regulated devices.

US vs EU configurations. Next-gen exploration alongside the current release. A three-way merge engine with explicit conflict resolution on every node type.

Closed loop

Post-market signals reach the next revision.

A field complaint links instantly to the affected BOM line, risk control, V&V protocol, and regulatory filing. The loop from complaint to design input — the one every team intends and few close — closes by construction.

The closed loop is not a workflow. It is a graph traversal.

Consolidates

What Avera replaces.

One platform where previously there were three — and the connective tissue that stitched them was spreadsheets.

ALM
Jama
Polarion
Helix ALM
PLM
Windchill
Arena
Teamcenter
QMS
Greenlight Guru
MasterControl
Veeva QMS
Connective tissue
Excel
Word
SharePoint
See the platform

A working demo, on your own device data.

Send us a sanitized slice of your current requirements and BOM. We'll show it to you as a graph.

Book a platform conversation