Launching today
CipherExplain
SHAP explanations on encrypted data — vendor never sees it
2 followers
SHAP explanations on encrypted data — vendor never sees it
2 followers
CipherExplain runs SHAP model explanations on encrypted inputs using fully homomorphic encryption. The vendor never sees plaintext features, model weights, or explanations. Built for regulated AI buyers (SR 11-7, EU AI Act, HIPAA). Python SDK.
Hey Product Hunt 👋
I'm Bader, solo founder of CipherExplain.
What it does:
CipherExplain runs SHAP model explanations on encrypted inputs using fully homomorphic encryption. The unusual thing is that I literally cannot see your data, your ML model, or even the explanations the service produces. The customer encrypts on their side, the server computes under encryption, the customer decrypts the result. Privacy by math, not by contract.
Why I built it:
every regulated AI team I've talked to has the same problem. Their own customers and auditors need model explanations (banks under SR 11-7, anyone under EU AI Act Article 13, healthcare under HIPAA). But sending features to a third-party "explainability API" usually breaks the same regulations the explanations were meant to satisfy. So the standard answer becomes "build it in-house" — which most small AI teams can't afford. CipherExplain is the third option: get the explanation, never expose the data.
My ask to the PH community: 1. If you're an ML engineer, CTO, or compliance lead at a regulated AI company — does this map to a real problem you have? Why or why not? 2. If you've tried FHE for ML and gave up — what killed it for you? I want to learn from your scar tissue. 3. If the latency (~72s/call) feels too slow, what number would make it useful for your use case?
Demo + SDK: cipherexplain.vaultbytes.com GitHub examples: github.com/VaultBytes/cipherexplain-examples DMs + email open: b@vaultbytes.com I'll be here all day answering anything — technical (CKKS params, boundary bootstrap, OpenFHE production gotchas) or product (positioning, pricing, roadmap). Thanks for any feedback 🙏