AI Evaluation Intern - Qriton HLM

About Qriton

Qriton Technologies develops HLM (Hopfield Language Models) - an inspectable, energy-based architecture for high-stakes structured data. HLM uses polynomial Hopfield cores with modality-matched frontends (language, vision, audio, sensor/edge). Knowledge is stored in discrete, addressable attractor basins that can be inspected, captured, injected, or surgically edited via Energy Language without full retraining.

Decisions expose uncertainty, evidence paths, convergence state, and replayable audit records. Public checkpoints and baseline validation numbers are available at hlm.qriton.com/status

The company is research-first and actively collaborates with universities on energy-based models, associative memory, and interpretability.

Role

You will join the research team to extend and rigorously stress-test the evaluation framework for HLM checkpoints and Energy Language operations. The internship focuses on developing defensible metrics for accuracy, calibration, uncertainty, convergence stability, and post-surgery basin integrity in a novel architecture where standard benchmarks are insufficient.

You will work with real checkpoints, contribute to the public validation surface, and help define how basin-centric models should be evaluated.

Responsibilities

Extend the existing evaluation harness to cover new metrics for energy-based models (basin stability, edit success rate, causal consistency of injected/removed concepts, energy landscape properties).

Design and run experiments quantifying calibration (ECE, Brier, selective prediction), OOD separation, and convergence behavior across language, vision, audio, and edge tiers.

Analyze model failures and trace issues to frontend geometry, core Hopfield dynamics, or specific attractor basins. Produce internal evaluation reports and contribute to public technical documentation and the Validation page. Collaborate on novel evaluation approaches that better capture the properties of polynomial Hopfield / energy-based systems.

Required

Currently enrolled in a Master’s or PhD program (or strong Bachelor’s with substantial research experience) in Computer Science, Machine Learning, Statistics, Physics, or a closely related field.

Strong Python + PyTorch proficiency demonstrated through research projects or substantial codebases.

Solid command of probability, statistics, and modern ML evaluation methodology (calibration, robustness, ablation design, statistical significance).

Strong foundation in linear algebra (eigenvalues, SVD, matrix decompositions, fixed-point analysis) and familiarity with statistical mechanics concepts relevant to energy-based models (energy landscapes, attractor dynamics, basin stability).

Experience extending existing research codebases rather than building from scratch.

Clear technical writing in English and comfort with Git/Linux/containerized workflows.

Primary Baseline Evaluation Framework

Our current evaluation framework consists of custom PyTorch harnesses built on top of the open-source qriton-hlm library. It includes tools for basin surveying, energy computation, post-edit validation, and modality-specific metrics. Interns are expected to integrate new experiments and metrics into this existing framework.

Working Arrangements, Compute Resources & Academic Collaboration

Remote Setup & Synchronous Overlap

The role is fully remote within the EU and primarily asynchronous. To ensure effective mentorship and collaboration, we maintain core synchronous overlap of approximately 4 hours per week (typically Tuesday and Thursday mornings in CEST, exact recurring slots agreed during onboarding). The remainder of the work is async with clear documentation and written updates.

Publication Policy

Novel evaluation metrics, methodologies, analyses, or insights developed during the internship are encouraged to be published. We support open research and will collaborate with you on timing, authorship (including company co-authorship where substantial), licensing, and venue. Any proprietary model weights, internal data, or non-public implementation details remain confidential per mutual agreement.

University Co-supervision & Thesis Integration

We are explicitly open to structuring the internship as part of a Master’s thesis, research project, or university collaboration, including formal co-supervision arrangements with your academic advisor. Please indicate this interest in your application or initial discussions. Scope, IP terms, deliverables, and timelines will be agreed jointly with you and your institution.

What We Offer

High-agency research work on a genuinely novel architecture with public checkpoints and visible validation impact.

Close, regular mentorship from the founding technical team.

Dedicated compute budget and the opportunity to influence core evaluation methodology for basin surgery and energy-based models.

Competitive monthly stipend aligned with strong EU research internship standards.

Flexible hours compatible with academic commitments.

Strong recommendation letter and potential pathway to future full-time or extended research roles.

Support for academic output and thesis integration.

Apply

Email [email protected] with subject: “AI Evaluation Intern Application - [Your Full Name]”

Include: CV or LinkedIn.

One artifact demonstrating rigorous evaluation or experimental work (GitHub repo with evaluation code + analysis, technical report/PDF from a project, or 1–2 page critical paper review focused on methodology). Add 2–4 sentences on what you learned and what you would improve.

One paragraph (3–5 sentences) on why evaluating HLM’s energy landscapes and basin surgery specifically interests you (rather than generic ML evaluation or interpretability).

We do not require a cover letter. Strong artifacts showing clear experimental thinking and technical depth will be prioritized.

Timeline: Rolling applications. We aim to respond within two weeks. Promising candidates will be invited to a short technical discussion focused on evaluation design and research thinking.

This internship is designed for high-caliber students who want to make a concrete contribution to how a new class of inspectable, editable AI models is rigorously validated. We look forward to your application.