ePrivacy // compute on locked data
group Trazeus
// a question

Can the cloud add up your data without reading it?

Hospitals want statistics. Normal analytics has to unlock every record first.

// the problem

Locked on disk. Open the moment it is used.

To compute, the server decrypts every record.

Insiders can read it in memory.

Hacked hosts and malware can scrape it.

Removing names does not hide people.

// our approach

Compute on data that stays locked.

The cloud adds ciphertext. Only the hospital can read the total.

// 01 · encrypt

Locked before it leaves the hospital.

A fresh key for every campaign. Names and IDs never leave.

e-privacy.seniru.dev/campaigns
A screenshot of the ePrivacy app
private key stays here identifiers dropped
// 02 · add blind

The cloud adds. The hospital opens.

Lock
→
Add blind
→
Open total
// 03 · under the hood

Multiply the locks, add the values.

Paillier encryption is additive. The cloud never needs the key.

hospital Enc(12), Enc(30)
the cloud multiplies them
Enc(12) × Enc(30) = Enc(42)
// 04 · verify

Five checks before anything opens.

Signature, log, input root, record count and a local recompute.

e-privacy.seniru.dev/analysis
A screenshot of the ePrivacy app
a cheating cloud is rejected then it decrypts
// 05 · audit

Anyone can audit it.

A public, signed log. An auditor needs no key and no trust.

e-privacy.seniru.dev/board
A screenshot of the ePrivacy app
runs in your browser four public checks

Encrypted. Added blind. Checked by anyone.

Ask us for a live demo →

scan to try it on your phone

// how it works · 1 of 3

From key to checked result.

01Keys

A fresh Paillier keypair for each campaign.

02Encrypt

Identifiers dropped. Every measure encrypted.

03Receipt

The cloud signs a Merkle root over every record.

04Add

Grouped sums computed on ciphertext.

05Verify

Five checks, then decrypt and chart.

// how it works · 2 of 3

Four facts, and how we know.

540,000 records

Ten departments, fully synthetic.

Paillier 2048-bit

Choose 1024 or 2048 per campaign.

~1 GB, not ~617 GB

Paillier against CKKS, 2 million records.

Groups under 10 held

No individual can be singled out.

Storage figures come from our own benchmark. No real patient data is used.

// how it works · 3 of 3

The parts, and how they fit together.

ciphertext stores reads Hospitalholds the key,encrypts, decrypts Cloudadds ciphertext,signs every receipt Auditorre-checks it allin the browser Bulletin boardsigned, hash-chainedpublic log