Arlington, VA · Open to AI Safety & Governance Roles

AntoineSavelli

AI Safety and Governance | Controls for LLM and Agent Systems in Financial Services

I build the controls that keep AI systems safe to run and auditable, and I find the errors that make results look better than they are.

I architect systems, direct AI coding agents to build them, and verify their output with tests and CI. The controls are the evidence: look-ahead scanning, a sizing-order check and a golden-regression baseline enforced in CI, ask-first gates on the agents, and an evaluation harness that enforces pre-registration. Before that, five years in research, where I tested an LLM pipeline against pre-coded data before it replaced a vendor contract.

See the Proof Get in Touch

Portfolio

Paramo

Founder and Engineer. A paper-stage system with live-grade architecture. It has not traded real capital and makes no performance claims. The work to look at is the controls: look-ahead scanning, a sizing-order check and a golden-regression baseline enforced in CI. I architect it and direct AI coding agents, and tests and CI verify their output.

View

pruvera

An evaluation harness that tests whether instruction changes help local coding agents. It plants defects, judges results with issue-clustered statistics and enforces pre-registration. Its clearest finding: file protection came from the permission layer, not from model behavior. Designed and directed by me, built with AI coding agents and verified by tests. Public since October 2026.

GitHub

AI Implementation in Action

Data Analyst, Operations at SimpsonScarborough. Three case studies: an LLM pipeline tested against pre-coded datasets before it replaced a $50k contract, AI-assisted automation of data preparation, and a survey-to-BigQuery pipeline feeding a live Tableau dashboard. Each one documents the problem, the fix and a measured result.

View

torra

A local agent designed so the model never receives a shell: what it proposes goes through five gates before anything runs. In progress and not yet public.

In progress

Resume

Five years of research and operations. Every layer of the stack. The full picture.

Download

Academic Credentials

B.S. Applied Psychology ↗ / Minor, Applied Statistics ↗

Cognitive Profile ↗

Supplement. Johnson O'Connor aptitude assessment, March 2026: Inductive Reasoning and Foresight at the 99th percentile, Analytical Reasoning at the 90th. Full report on request.

References 🔒

Available upon request. Contact Antoine for access.

Skills & Tools

AI Safety & Governance

Agent Permission Gates / Evaluation Design / Pre-registration / Issue-Clustered Statistics / CI Guardrails / Look-ahead Controls / Audit Trails

AI & Automation

LLM Integration / Prompt Engineering / Workflow Automation / AI Tool Evaluation

Data Infrastructure

Pipeline Design / REST APIs / ETL Processes / Google BigQuery / Tableau / SPSS / Data Cleaning

Research

Survey Design / Sample Design / Market Research / Thematic Synthesis / Statistical Analysis / Presenting Findings

Agentic Development

Directing AI Coding Agents / Test & CI Verification / Python, R, SQL & Excel VBA (AI-Assisted) / SPSS Syntax

Platforms & Tools

Excel (Advanced) / Qualtrics / Microsoft Access / GitHub / Google Suite / Microsoft 365

Cognitive Profile ↗

Inductive Reasoning / Foresight / Analytical Reasoning / Systems Thinking / Independent Problem Solving

Get in Touch

Phone

408.833.3360

antoine.savelli@outlook.com

antoine-savelli

GitHub

antoinesavelli

Open first to AI safety, governance and compliance roles, especially in financial services: AI governance, AI risk, model risk and AI evaluation. Also open to AI implementation and research operations work. Reach out directly or use the form.

Arlington, VA · Open to Remote