Jesús Tabares Montilla
Founder of Seetrex. I work on Deterministic Intelligence: inference that is verifiable, traceable and controllable, through runtime control, security and auditability.
Work
-
Seetrex Warden: an external obligation gate for a long-horizon LLM agent. A pre-registered pilot (n = 1 per arm) on the Ground Control scenario of CivBench (arXiv:2609.02459)
An MCP proxy that keeps a ledger of obligations with due dates and rejects
end_turnwhile one is overdue; it never says what move to make. Pilot on CivBench, one game per arm. -
Laimark: Gains and Structural Limits of Self-Generated Curricula in Reinforcement Learning from Verifiable Reward
On HumanEval with Qwen3-8B, a problem set generated by the model itself gives measurable gains under GRPO and hits structural limits; the paper says what those limits are.
-
Monolith: Differentiable EML Trees for Symbolic Regression via Gradient Descent on a Universal Operator
EML trees as trainable PyTorch modules: recovers 7/7 elementary functions with ≤24 parameters by gradient descent on one universal operator, with baselines against PySR and an MLP.
-
The Geometric Blind Spot of Perplexity: When Low Loss Hides Out-of-Distribution
Perplexity misses out-of-distribution inputs where the model is fluent but the task does not match; a geometric view of the hidden states catches them (LLaMA-3-8B, Mistral-7B).
-
KuraFormer: Oscillatory Dynamics as Parameter-Efficient Adapters for Pretrained Transformers
Version 2 corrects the claims of the original preprint: a static-adapter ablation shows the Kuramoto oscillatory dynamics add nothing over a bottleneck adapter (54.1 % vs 53.9 %).
-
Runtime Compute Control: Explicit Inference-Time Computation Management for Large Language Models
A post-training execution layer that exposes one compute-ratio parameter to control which paths run at inference, validated on Qwen2.5-3B with a C++ runtime.
Projects
- Seetrex, a deterministic inference engine in Rust with no ML dependencies. Its obligation gate for LLM agents, Warden, is the subject of the pilot above.
- SEEQuant, interactive quantum software. Its first app, Syndrome, makes quantum error correction tangible in the browser: a rotated surface code whose decoder is validated against PyMatching and Stim.
Contact
jesus@tabares.eu. ORCID 0009-0001-5686-6765. Code at github.com/seetrex-ai.