I build production software and evaluate machine-learning systems with an emphasis on traceable evidence, explicit uncertainty, and safe failure. My public work focuses on reproducible research and practical developer tools.

The projects below are the best entry points into my work. Research repositories publish methods, aggregate evidence, and limitations; they do not redistribute licensed datasets, subject-linked records, or private product logic.

rppg10-eval-harness
A reproducible evaluation harness for camera-based physiological-signal research, with ground-truth scoring, documented limitations, and data-rights boundaries.

rppg-skin-tone-equity
An aggregate fairness evaluation for camera-based physiological sensing, with limitations reported alongside the result and no gated corpus redistributed.

eeg-affect-honest-negatives
Pre-registered EEG experiments that publish negative results when held-out evidence does not support the claim.

system-brain-mcp
Read-only MCP tools for inspecting software systems and returning evidence-linked answers. Missing or degraded evidence produces fewer claims, not invented certainty.

hf-papers-mcp
A pure-Node MCP server for public Hugging Face Papers search, daily digests, metadata, and citations.

obsidian-icloud-mcp
An MCP server for Obsidian vaults that detects unavailable iCloud files before reading and reports incomplete searches explicitly.

claude-honest-engineering-skills
Reusable engineering procedures for validation, security triage, stub detection, and evidence-based release decisions.

I am building Quantara, an applied-AI product with a mobile app, machine-learning systems, backend services, and analytics. Product architecture, operating thresholds, customer workflows, model-selection decisions, and private research direction are not published here.

Repository social artwork

• Validate the leaf behavior, not just the surrounding architecture.
• Separate measured outcomes from model-generated guesses.
• Report coverage and limitations with every result.
• Treat abstention as a valid answer when the evidence is insufficient.
• Keep gated datasets, customer data, internal identifiers, and operating thresholds private.

I am open to collaborations in physiological-signal validation, affective computing, reliable agent tooling, MCP systems, and evidence-led machine learning.

quantaraliving.com · hello@quantaraliving.com