NEVER SHIP ON FRIDAYS
MENU
WORK / LAB INDEX
THE MACHINERY BEHIND THE THESIS

PROBLEM → EXPERIMENT → SYSTEM → EVIDENCE

The laboratory is not a metaphor.

Project OVI is the broader research and systems program. AVA is its originating continuity system; OVI Cards is its independently published protocol. This index covers that laboratory and selected technical receipts. Commercial and domain products are catalogued on Andrew's working profile.

Open the working profile →

PROJECT / 01

AVA

Relational continuity should survive the model underneath it.

Enter the AVA research system →
[working]
SYSTEM

The originating research system within Project OVI, exploring persistent identity, relational continuity, user-owned memory, model independence, and verifiable behavior.

[disclosure]

Public research record

A deliberately bounded account of the research questions, chronology, observable properties, and evidence posture.

[runtime]

Runtime evidence

Contemporaneous records support continuity-related behavior influencing later interactions. Mechanism withheld.

[filing]

Initial patent filing

A U.S. provisional application records protected work at the filing boundary. Patent pending.

PROJECT / 02

Hippo

Memory needs an owner, a lifecycle, and a way to prove it came back.

[working]
SYSTEM

A sovereign memory service for durable handoffs, scored retrieval, provenance, and explicit retention behavior.

[live]

Sovereign memory service

Durable scored cards, explicit retrieval, service health, and recovery-oriented memory operations.

[postmortem]

Eviction incident

A threshold mismatch destroyed most of a live corpus and produced a concrete recovery and safeguard trail.

PROJECT / 03

OVI Cards

An agent action without a receipt is merely a confident story.

Install and test OVI Cards →
[published]
SYSTEM

Object-verified interaction contracts that bind claims and actions to inspectable evidence.

[repo]

OVI Cards 0.1.0

Installable MIT-licensed protocol: five typed card classes, JSON Schema and Pydantic validation, fifteen semantic checks, builders, examples, tests, and an MCP server.

Open public receipt →
[bench]

Reproducible context benchmark

The repository includes its benchmark script and reports a 69.5% median reduction across seventeen task-oriented samples. Independent replication is still welcome.

Open public receipt →
PROJECT / 04

Corlia

Applied AI earns trust by surviving contact with an actual workflow.

Visit the closed-beta product ↗
[closed beta]
SYSTEM

A DISC-aware AI field coach for real-estate professionals, carried from product architecture through authentication, persistence, resilient streaming, model portability, billing, storage, access controls, testing, and deployment.

[live]

Closed-beta product

A deployed, invitation-only AI coaching product with public landing, pricing, privacy, and terms surfaces.

Open public receipt →
[source]

Full-stack implementation

Authentication, PostgreSQL persistence, resilient streaming, model-provider abstraction, billing architecture, file storage, onboarding, and beta access controls operate behind the public surface.

PROJECT / 05

R ███████████

[withheld]
PROJECT / 06

Friday Runtime

The agent should outlive any single model, machine, or interface.

[working]
SYSTEM

A distributed execution fabric for routing work across local and cloud models while preserving authority boundaries.

[source]

Distributed runtime

Model and machine routing with sensitivity boundaries, health state, and recovery mechanisms.

PROJECT / 07

Brain Trust

Specialists are useful only when disagreement and authority remain visible.

[in progress]
SYSTEM

Coordinated specialist intelligence with explicit provenance, role boundaries, and conflict handling.

[unverified]

Specialist coordination

Active work on role-scoped reasoning, provenance, and visible disagreement.