ImageWeightAudit

Who it's for

DevOps/ML infra engineers and security engineers who today cope by running ad-hoc grep/find commands, trivy/clair scans, or manual image inspection before pushing images.

The problem

Time and operational risk: accidental image bloat, IP/license exposure, large registry egress costs, and potential leaks that trigger legal or security incidents; current tooling misses large binary blobs (weights) and offers no signed proof for audits.

How to build it

Command-line tool + optional small web UI for aggregated reports and signed manifests; integrates into CI (GitHub Actions/GitLab CI) and container registries via a simple webhook check.

How it makes money

Who pays: small-mid engineering/security teams or platform teams; Pricing: $20–$100/month per team for hosted aggregation + signed-manifest storage, or $200–$1,000 one-time for an on-prem binary/license for regulated orgs — they pay because registry transfer/egress costs, legal risk and audit time easily exceed those fees and there is no free tool that gives signed, tamper-evident manifests plus a rebuild script.

Why it doesn't exist yet

Incumbent container scanners focus on package-vuln/secret detection and don't reliably fingerprint arbitrary large binary blobs (model checkpoints) or capture license metadata tied to weight files; infra teams lack a tiny, fast tool that (a) detects big-weight patterns inside layered images, (b) outputs a signed manifest for audits, and (c) generates deterministic strip-and-rebuild remediation automations — a niche incumbents ignore because it's low-volume and ML-specific.

First users

Early users: MLEs and devops at startups shipping LLM-powered features who recently faced 'exfiltrate-your-weights' headlines and want quick auditability; security teams at AI consultancies needing signed evidence for client audits; two to three infra engineers will adopt it because it's small, CI-friendly, and fixes a visible risk in minutes.

Build size

1 person x 8 weeks to build CLI + CI integrations + minimal hosted manifest store; add 1 designer/devops tester for a total of 2 people x 12 weeks to harden registry/webhook flows and the optional UI.

Biggest risk

A major container-scanning vendor (Snyk/Anchore/Trivy) adds fast, signed-large-binary-detection and CI remediation with free tier integrations, commoditizing the niche; or container registries add built-in weight-detection scanning and signed manifests as a native capability.

Conditions for a hit (all 3 required)

• Detect output: Given a container image reference or tarball, produce a JSON manifest within 30s listing every embedded file >10MB with file path, SHA256 hash, file size, and a best-effort license tag (from nearby package metadata or file heuristics).
• Remediation output: Emit a deterministic 'strip-and-rebuild' shell script that removes listed weight files (or moves them to a configured registry/host) and rebuilds a layer-equivalent image in under 60s for small images (<=1GB), plus a CI-friendly return code for gating pushes.
• Auditable signature: Upload or locally-sign the produced manifest producing a tamper-evident signature (PKCS7 or similar) that can be verified later and includes timestamp and scanner version so security/audit teams can prove a pre-push scan occurred.

How it's judged (in 6 months)

GitHub 500 stars or Product Hunt top 10(judgment date 2027-03-24)

AI self-confidence 55/100 — self-reported likelihood of meeting the criterion, not a business success rate