Python · PyTorch · Weights & Biases

• Built a ray tracerfrom scratch.
• Implemented ResNet-34 from scratch and used it as a frozen feature extractor on CIFAR-10 (trained weights copied in, only the final linear layer optimized).
• Implemented SGD, RMSProp, and Adam from scratch.
• Implemented backpropagation in NumPy.
• Implemented and trained an autoencoder, a VAE, and a GAN.
• Implemented and trained a transformer from scratch. Also implemented different sampling methods like greedy, beam, top-p, top-k and so on. Currently working on mechanistic interpretability.

Rust · Docker

• Implemented the Chaum-Pedersen zero-knowledge proof from scratch in Rust: prove knowledge of a secret without revealing it.
• Modeled the full 3-step ZK handshake: commitment t = gʳ, challenge c, response s = r + c·x, grounded in the hardness of the discrete-log problem.
• Engineered modular prover / verifier with verification gˢ ≡ t·yᶜ and comprehensive unit tests.
• Built a containerized gRPC server exposing the protocol over a client-server architecture, deployable via Docker.

Rust

• Implemented ECDSA signing and verification from scratch on secp256k1, the curve Ethereum uses for keypair generation.
• Derived Ethereum addresses and uncompressed public keys from raw 256-bit private keys via elliptic-curve point multiplication.
• Built a CLI that signs any file type (images, binaries, documents) by hashing its contents into a detached signature file.
• 3-command workflow: generate-keypair, sign, verify , practical for real-world file integrity and authenticity.

Rust

• Implemented the logic to derive an Ethereum / Bitcoin wallet address from a given private key.
• Implemented all curve operations and the underlying finite-field arithmetic.