FHE Accelerator Toolchain

This project focuses on turning algorithmic insights into practical tooling:

• an intermediate representation for CKKS-heavy compute graphs
• cost models for automorphism, key-switch, and NTT operators
• scheduling heuristics that target throughput under memory limits

The resulting prototype reduced kernel launch overhead and improved end-to-end inference consistency across multiple parameter sets.