Independent AI Safety Research · Production Evaluation

Precision systems for adaptive intelligence.

Research-grade AI safety tools for detecting instability, behavioral drift, and unsafe escalation before failure appears in model outputs.

LLM Evaluation · Inference-Time Stability · RAG & Agentic Systems · MCP Tooling

cos = 0.1524Neutral‑anchored median
0%Antipodal fraction
38.8%L/R coactivation
−1 tokenPre‑leak lead
6 / 6Constraint retained
4 / 6 LR‑action block
Inside the CRAD case study →
Core Thesis
The signal appears before the failure.
Model failure is not always a final answer. Sometimes it is a trajectory.
Click to reveal
TwoQuarks
Instability can be measured before it becomes visible.
TwoQuarks studies drift, divergence, and pre-critical structural change during inference. The work translates research signals into tools for LLM evaluation, monitoring, and operational AI safety.
Click to return

About & Development

Independent AI-safety researcher. I build the instruments that catch model instability before it turns into a visible failure.

What I actually do

I'm Jaime Ledesma — independent researcher running TwoQuarks out of Guadalajara. No lab behind me, no cluster. Just the work.

The whole bet is one idea: a model doesn't break all at once. Before the bad output, the internal state already drifts. Measure that drift and you catch the failure a step early — without touching weights, retraining, or rewriting the policy.

So I build the instruments that measure it. From the outside — PfV / Molecule, reading structural divergence straight off the API outputs. And from the inside — CRAD, reading the residual stream to flag a leak while the safety constraint is still switched on. Preprints, a Python package, an MCP tool, and a playground you can run yourself.

who: Jaime Ledesma — independent AI-safety researcher
based: Guadalajara, MX · no lab, no cluster
thesis: the signal shows up before the failure
methods: black-box PfV / Molecule · white-box CRAD
ships: 4 preprints · twoquarks (PyPI) · MCP tool · live playground
stack: Python · PyTorch · HF · APIs · MCP · RAG
status: open to AI-safety collaboration & roles