Algorithms Engineer, Differential Privacy at Oblivious

Bengaluru, karnataka, India -

Full Time

Start Date

Immediate

Expiry Date

16 Jan, 26

Salary

0.0

Posted On

18 Oct, 25

Experience

2 year(s) or above

Remote Job

Yes

Telecommute

Yes

Sponsor Visa

Skills

Algorithms, Differential Privacy, Python, Statistical Modelling, Privacy Accounting, SQL, Compiler Design, Machine Learning, Data Generation, Sensitivity Analysis, Static Analysis, Numerical Stability, Error Reasoning, Privacy Mechanisms, Noise Mechanisms, Data Collaboration

Industry

Data Security Software Products

Description

Ever wanted to join a vibrant young start-up? To tangibly change the world for the better? Oblivious builds privacy-enhancing technologies to help organisations unlock insights from sensitive data. We are recruiting an Algorithms Engineer to design and implement the core components of our differential privacy (DP) systems, including our Private Python runtime, DP-SQL engine, and synthetic data generator. This role requires translating mathematical theory into production-ready code. You will work on the fundamental challenges of making rigorous privacy guarantees practical and efficient. Who We Are: Oblivious is a start-up focused on enabling secure data collaboration through privacy-enhancing technologies. We were founded by two former PhDs in machine learning and cryptography from the University of Oxford who are on a mission to make privacy-preserving technologies the new norm across the industry. We are backed by some of the most well-respected VCs in Europe and the US, and we are putting together a core product and development team. You will get to build platforms that are leveraged by the largest financial institutions and telecoms companies in the world. Responsibilities Privacy Accounting & Mechanisms: Implement and analyse privacy loss accountants (RDP, zCDP) and their conversions to (ϵ, δ)-DP. Calibrate and apply noise mechanisms (Gaussian, Laplace) based on rigorous sensitivity analysis. Differentially Private SQL Engine: Develop algorithms for static and dynamic sensitivity analysis of relational operators. Build query rewriting logic to inject calibrated noise and manage a per-user privacy budget ledger. Compiler & Static Analysis: Use Python AST manipulation and static analysis to enforce a DP-safe execution environment, ensuring user-submitted code cannot leak private information. DP Synthetic Data: Implement and benchmark state-of-the-art algorithms (e.g., MWEM, PGM, PrivBayes variants) for high-dimensional synthetic data generation, analysing their privacy-utility trade-offs. Strong foundation in probability, statistics, and linear algebra. You must be comfortable with statistical modelling, proving bounds, and reasoning about error/variance. Proficiency in Python for scientific computing, including numerical stability considerations (e.g., floating-point precision, clipping, scaling). Demonstrated ability to translate mathematical concepts from academic papers or technical specifications into robust, well-tested code. Desirable Direct experience with differential privacy concepts or libraries (OpenDP, SmartNoise, TensorFlow Privacy). Knowledge of compiler design, abstract syntax trees (ASTs), or program analysis. Experience with machine learning, particularly with noise models, statistical learning theory, or generative models. Familiarity with SQL parsers or database internals. Private Health Insurance Paid Time Off Work From Home, with one required in-office anchor week every six weeks for deep collaboration and planning Training & Development Oblivious Software Limited is committed to equal opportunity for all. We may collect, store, and process relevant personal data as part of our candidate evaluation process in accordance with our privacy policy at https://www.oblivious.com/policies-oblivious/privacy-policy

Responsibilities

Design and implement core components of differential privacy systems, including a Private Python runtime and DP-SQL engine. Analyze privacy loss accountants and develop algorithms for synthetic data generation.