Towards a Standard Interface for Differential Privacy Systems with Database Interoperability
Coach Name
Jordi Bosch Garcia
EU Organization
DPella AB
Members
- Carola Compa
- Alejandro Russo
- Marco Gaboardi
- Elisabet Lobo
- Matthias Guissurarson
US Organization
Harvard University (OpenDP Initiative)
Members
- Michael Shoemate
Project Overview
DPDB tackles a major barrier to real-world adoption of Differential Privacy (DP): the lack of interoperability between DP systems and the SQL engines that power virtually all organisational data infrastructures. Most organisations store sensitive data in relational databases, but existing DP tools rely on custom runtimes that cannot be dropped into standard SQL workflows.
DPDB designed and implemented a database-agnostic compilation framework that converts DPella’s relational-algebra queries into optimized SQL, enabling privacy-preserving analytics directly where data already lives. The team built a DP interoperability protocol that safely injects noise through database-callable functions, validated the approach across SQLite, PostgreSQL, and MySQL, and demonstrated full semantic equivalence across engines.
This work provides a practical pathway for companies, hospitals, public agencies, and researchers to adopt Differential Privacy without replacing their database stack, accelerating safe data sharing and GDPR-aligned data analytics.
Methods and approaches
Database-Agnostic Compilation Framework
DPDB created a full compiler that translates DP relational algebra into SQL while preserving semantics and privacy guarantees. The process includes:
- mapping operators like projection, filtering, classification, partitioning and aggregations to SQL;
- injecting DP noise via a database-callable function (dpella_sample_random);
- applying systematic SQL optimizations (merging filters + measurements, removing unnecessary projections).
Cross-Engine Interoperability Protocol + Differential Testing
A lightweight protocol enables SQL engines to call Haskell-based DP functions safely, with implementations for SQLite (embedded), PostgreSQL and MySQL (server engines).
The team validated correctness using zero-noise differential testing, ensuring identical outputs between reference and SQL-based implementations.
Key Achievements
Interoperability Report mapping integration routes (UDFs, C-extensions, FFI) for SQLite, PostgreSQL, and MySQL.
Three complete interop examples demonstrating DB ↔ DP runtime communication.
Technical report detailing methodology, optimizations, and privacy semantics.
10 compiled DP queries for SQLite and PostgreSQL, with optimization notes.
Semantic-preservation tests across engines with differential testing methodology.
Prototype interoperability protocol, demonstrated in a public video.
Public project website and contact mailbox for community reuse.
Five joint EU–US mentoring/workshop sessions strengthening alignment with the OpenDP community.
Impact & Results
Scientific Impact
DPDB advances the state of the art in integrating Differential Privacy into mainstream data systems. It provides:
- formalized methods for DP-to-SQL compilation,
- reproducible testing showing semantics preservation,
- open technical reports and public artefacts that deepen transparency.
Economic / Industrial impact
The framework significantly reduces deployment friction by enabling DP within existing SQL environments, lowering integration costs for regulated sectors such as health, finance, and public administration.
It allows organisations to adopt privacy-preserving analytics without replacing databases, reducing both cost and energy use.
Societal Impact
DPDB supports GDPR-aligned data sharing, enabling institutions to extract insights from sensitive data while respecting citizens’ privacy rights.
The project strengthens public trust in digital analytics and helps ensure that data-driven innovation aligns with privacy protections.
EU–US Collaboration
The partnership with Harvard’s OpenDP initiative produced:
- co-design of testing methodologies,
- technical review of semantics and privacy behavior,
- joint workshops and community presentations,
- alignment toward future standardization of DP systems.
This collaboration is ongoing and expanding into adjacent research topics.
Publications and Open-Source Contributions
- GitHub repository: https://github.com/dpella/interop-example
Includes:
- interoperability report,
- open technical report,
- compiled SQL examples,
- differential testing harness,
- PostgreSQL/MySQL/SQLite integration code,
- demo video.
- interoperability report,
- Project website: https://dpella.github.io/interop-example/
- Conference/community presentations: OpenDP Community Meeting 2025 and Eyes-Off Data Summit 2025.
Future directions