What I Build
This page is where I put the work that doesn’t fit neatly into a paper—the models, tools, and systems I’ve built to make LLMs more capable, more secure, and easier to evaluate.
Most of it comes out of Cisco Foundation AI, plus an open-source deep-RL framework for wireless systems that the community still uses.
⚙️ FAPO: Fully Automated Prompt Optimization of Multi-Step LLM Pipelines⌗
Co-Creator · 2026
A framework that lets Claude Code autonomously optimize multi-step LLM pipelines—diagnosing failures, applying scoped prompt edits, and escalating to structural changes only when needed; outperforms the GEPA baseline in 15 of 18 model–benchmark comparisons.
🤗 Foundation-Sec-8B-Reasoning⌗
Core Model Developer · 2026
Built and released the reasoning-tuned variant of Foundation-Sec-8B; further improves Foundation-Sec-8B-Instruct on complex, multi-step cybersecurity tasks.
🛡️ Latent-Space AI Firewall⌗
Creator · 2025
An LLM guardrail that operates on the hidden states of Foundation-Sec-8B-Instruct; outperforms text-level filters (e.g., Llama Guard 3-8B) at detecting malicious inputs and adversarial prompts.
🤗 Foundation-Sec-8B-Instruct⌗
Core Model Developer · 2025
Built and released the instruction-tuned (+RLHF) variant of Foundation-Sec-8B; achieved state-of-the-art results on cybersecurity benchmarks and instruction-following tasks.
🤗 Foundation-Sec-8B⌗
Core Model Developer · 2025
Developed and released Cisco’s cybersecurity-focused foundation model (pretrained transformer); achieved state-of-the-art performance on cybersecurity benchmarks versus comparable-size models.
🧪 FAITH: Foundation-AI Testing Hub for Cybersecurity⌗
Co-Creator · 2024–2025
Open-source evaluation harness for benchmarking language-model knowledge and capabilities in cybersecurity.
📡 Deep RL for Joint Beamforming and Phase-Shift Optimization (RIS-MISO)⌗
Creator & Maintainer · 2021 · ⭐ 208
Python framework for optimizing RIS-assisted multiuser MISO systems via deep RL, targeting 6G wireless settings.
[code]