I design defensive auditing systems at the intersection of generative AI and the measurement of structural coherence.
My primary system, SAS (Symbiotic Autoprotection System), is a forensic auditing layer that detects structural hallucinations in AI outputs—whether in text or code. It operates using an empirically calibrated coherence threshold of κD = 0.56, with a structural collapse threshold of κR = 0.15, and employs a multi-metric tribunal consisting of 7 deterministic, non-LLM modules (CPU-based).
This work is currently under peer review at Qeios (DOI 10.32388/T420IV), archived on OpenAIRE, hosted on Zenodo/CERN, legally registered in Argentina (TAD EX-2026-18792778), and timestamped on the Bitcoin blockchain via OpenTimestamps.
In production, it includes a public API (sas-api.onrender.com) with a `/v2/audit` endpoint, official SDKs available on PyPI (`sas-client`) and npm (`sas-js`), and a public Observatory featuring ISI results across 5 datasets.
I have also published the Symbiotic Architecture Protocol (SAP)—a formal security co-architecture protocol for multi-agent AI systems—which carries its own independent DOI.
Tech stack: Python, FastAPI, forensic traceability, reproducible research, and API hardening.
Current research focus: detection of structural hallucinations, temporal interaction auditing, defensive AI auditing, and backend infrastructure with hardened surfaces.