CLI coding agent for solana programs. writes anchor, builds, debugs-fully autonomous.

long-term memory architecture for LLM agents. achieved 67% SBERT similarity, 49% ROUGE-L, 48% F1 on 199 QA tasks.

deployed qwen3-4B-instruct using vLLM on modal A10G GPUs. 337 tokens/sec throughput, 6.7s P50 latency.

texas-hold-em arena where LLMs bluff, bet and banter in character.

a containerized tool for scanning and reporting vulnerabilities in local or docker-based applications.

a character-level language model inspired by qwen3, implemented and trained on shakespeare dataset.