SoloScale AI OS · Product case study

From AI-generated engineering work to verifiable capability.

AI agents can ship software faster than an operator can absorb it. SoloScale keeps engineering evidence, personal mastery, and public claims separate.

Watch the 85-second demo View the evidence repository

Hero Demo

Build once. Verify once. Compound everywhere.

85 seconds · 1920×1080 · captions burned in · no live model or network activity in the recorded verification path

Problem

Software can exist before the operator can defend it.

AI coding changes the bottleneck. The question is no longer only “Can this be built?” It is also “What was actually proven, and what can the operator independently explain?”

Approach

Three truth domains stay separate.

Engineering evidence — project, code, test, and decision anchors.
Personal mastery — independent human explanation and demonstration.
Public claims — only what survives evidence and ownership gates.

Workflow

Evidence first. Stronger claims later.

Selected Git project → evidence refresh → Learning Case → source/code/test anchors → evidence graph → deterministic receipt → human Explain step

Three public-safe receipts

Grounded, traceable, and bounded.

SoloScale Work Context showing a sanitized local Git project and evidence refresh control. — **01 · Grounded in real work.** Only an explicitly selected local project is read.

SoloScale interactive evidence graph connecting a project to sources, decisions, and capabilities. — **02 · Traceable engineering evidence.** A capability remains connected to its source and code/test anchors.

SoloScale deterministic receipt beside readiness and resume-eligibility boundaries. — **03 · Verified does not mean mastered.** Engineering truth is verified while interview readiness and resume eligibility remain separate.

What this demo proves

Packaged local macOS product
Explicit Git evidence refresh
Traceable source/code/test relationships
12 structured artifacts, 8 anchors, verified graph integrity
Human-in-the-loop mastery boundary

What it does not prove

Autonomous or personal mastery
Live model reasoning in this recording
Tests executed by the recorded run
Production customer adoption

Captured verification receipt

Deterministic by design.

12structured artifacts

8code + test anchors

0model calls

Nonetwork used

Tests executed by this recorded run: No. Broader build verification is a separate receipt.

2026-10-02 · Local release closeout

SoloScale 0.4.2 — verified delivery scope.

The current release keeps source facts, generated drafts, personal mastery, and public claims separate.

Local verification: 720 tests passed; 10 explicit skips for optional PostgreSQL / separate BuildLog integration. Ruff, strict mypy, and wheel/sdist builds passed.
Desktop candidate: 0.4.2 / build 9 installed locally from the same clean source SHA as its backend; authenticated native startup and packaged generation/download smoke passed. This is an ad-hoc local candidate, not a notarized public binary.
Chinese resume evaluation: one real, operator-selected resume and JD; two source-supported project edits in the final draft, inspected in a one-page native PDF. Three bounded local model attempts, zero paid calls; earlier attempts failed quality. Human application approval remains required.
Service evidence: a separate synthetic template evaluation ran 20 tasks across two local workers and checked recovery into human review. It does not establish paid-call exactly-once behavior or production capacity.
Current boundary: Resume Intelligence remains a preflight. The Showcase is static; AWS / a public live Resume API are deferred.

Inspect the delivery gate and reproduction steps · Review the source integration

Source release: 787264f · Version 0.4.2. Final main CI passed on Python 3.11 and 3.12. Full source provenance is recorded separately for the desktop candidate.

The 85-second recording above is the original frozen Hero Demo. Its zero-model-call receipt describes that recording, separately from the real Chinese evaluation.