Open source projects and contributions focused on cryptography, privacy-preserving technologies, and high-performance computing.
Pure Rust implementation of TFHE for boolean and integer arithmetics over encrypted data. Core cryptographic library with Rust, C, and WASM APIs. Contributor to this major open-source project by Zama, implementing programmable bootstrapping and advanced FHE features.
Decentralized application (dApp) for posthumous encrypted messages using smart contracts on FHEVM. Features check-in mechanism, grace period, encrypted message storage, and delivery system. Deployed on Ethereum Sepolia testnet as a proof-of-concept.
Encrypted blockchain chat rooms powered by Zama's FHEVM. End-to-end encrypted messaging with wallet-based identity, allow lists, and room management. Fully on-chain chat rooms deployed as smart contracts on Ethereum Sepolia.
Shell tool that parses Criterion.rs benchmark output under target/criterion/ and prints latency, throughput, confidence intervals, and regression deltas in color-coded terminal tables—optional JSON export and filtering for Rust performance workflows.
GPU-accelerated AES-128 on fully homomorphic ciphertexts using CUDA and TFHE-rs: full 10-round pipeline (SubBytes via Boyar–Peralta S-box, ShiftRows, MixColumns, key expansion) with multi-GPU support and batch evaluation—plaintext is never decrypted during the homomorphic computation.
Cyclotomic polynomial arithmetic library extracted from cuYASHE. CUDA-based implementation for high-performance polynomial operations used in RLWE-based cryptographic schemes. Provides exceptional GPU performance with a simple, maintainable API.
GPGPU implementation of the leveled fully homomorphic encryption scheme YASHE. Employs CUDA, the Chinese Remainder Theorem and the Fast Fourier Transform to achieve speedups of 6-35x for homomorphic multiplication compared to CPU implementations.
Framework for searching encrypted databases using order-revealing encryption and homomorphic encryption. Built on top of MongoDB with proof-of-concept implementation of Netflix Grand Prize predicates.
Efficient implementation of the CKKS homomorphic encryption scheme on CUDA using hierarchical Discrete Galois Transform. Achieves up to 3.6x faster homomorphic multiplication than state-of-the-art.
Efficient implementation of the BFV homomorphic encryption scheme on CUDA using hierarchical Discrete Galois Transform. Achieves up to 3.6x faster homomorphic multiplication than state-of-the-art.
Umbrella repository for GPU-accelerated CKKS homomorphic encryption components: aoa-dgt (Discrete Galois Transform path, GPL-3.0) and aoa-ntt (NTT-based newckks stack, MIT)—CUDA/C++ with NTL/GMP, hierarchical transforms for encrypted polynomial arithmetic.
C++/CUDA prototype that trains logistic regression on CKKS-encrypted features using the AOA libraries: homomorphic forward/backward passes on GPU, encrypted and plaintext MNIST and tabular runs (e.g. heart disease), scripted batch benchmarks for method comparison.