DrantechLabs Get in touch

Research with
something at stake.

Better performance. More control. Fewer fragile systems. These are proposed investigations, with a practical question behind each one.

These directions describe what I want to investigate, not completed breakthroughs or products for sale.

Why these problems

Private AI systems

Near-term investigation

Useful intelligence. Under your control.

Local models, permission-aware knowledge retrieval, and assistants that can show their sources and ask before acting.

Opportunity & first test

Where it could be useful

A private assistant that a small organization can actually maintain.

What I would test

Measure answer quality, permission isolation, latency, and support time against a simple search baseline.

The starting point

Local inference and document retrieval already exist. The question is whether a complete system can earn trust in everyday work.

Efficient computation

Near-term investigation

More useful work from the same hardware.

Model memory, inference scheduling, and data-specific compression. Performance improvements that survive real workloads.

Opportunity & first test

Where it could be useful

Lower operating costs and a longer useful life for existing hardware.

What I would test

Compare quality, memory, energy, and end-to-end time on held-out workloads. Count dictionaries, model weights, and decoding costs.

The starting point

Quantization, caching, and strong compression libraries are established tools. Any new approach needs to beat a well-tuned baseline.

Dependable automation

Near-term investigation

Systems that recover when reality interrupts.

Local voice, home and business automation, device coordination, and recovery from outages or incomplete information.

Opportunity & first test

Where it could be useful

Maintainable installations and repeatable automation products.

What I would test

Introduce network outages, stale sensors, ambiguous requests, and device failures. Check recovery, manual override, and unintended actions.

The starting point

Local voice and device control already work. Reliability across mixed devices and changing environments is the engineering question.

Verifiable research & software

Exploratory direction

Can someone else check the result?

Reproducible experiments, signed artifacts, explicit assumptions, and formal checks for narrowly defined claims.

Opportunity & first test

Where it could be useful

Evidence tools for developers, independent labs, and technical reviewers.

What I would test

Have another person reproduce a result from a versioned bundle. Check a small mathematical or software claim with an independent verifier.

The starting point

Proof assistants and artifact signing already exist. Useful integration and review workflows matter more than inventing another signature format.

Privacy & cryptographic migration

Exploratory direction

Make the trust boundaries visible.

Cryptographic inventories, practical migration testing, and privacy-preserving system design using reviewed standards.

Opportunity & first test

Where it could be useful

Clear migration reports and compatibility testing for smaller organizations.

What I would test

Inventory a permitted test environment, identify algorithm dependencies, and measure interoperability and resource costs.

The starting point

Post-quantum standards are already available. Finding dependencies and moving existing systems safely is the practical challenge.

Thermal & physical systems

Longer-term direction

Bring the measurement into the physical world.

Quiet compute, cooling, sensors, and repairable hardware. A path toward product engineering and, with the right partners, materials research.

Opportunity & first test

Where it could be useful

Validated cooling designs, monitoring tools, and useful specialist hardware.

What I would test

Measure temperature, acoustics, power, and serviceability on a controlled bench before making product claims.

The starting point

Start with testable thermal and mechanical problems. Novel materials require equipment, characterization, and specialist collaboration.

The frontier keeps moving.

Before calling a problem open, check the latest results, their assumptions, and their verification status. AI-assisted mathematics makes that habit more important.

My October 11, 2026 notebook entry looks at OpenAI’s recent mathematics release and what it means for choosing useful research.

Read the research outlook

Foundations still matter.

Mathematics, symbolic systems, automata, and information theory remain tools for the work. Materials science is a longer-term interest, with specialist collaboration and proper characterization needed before claims.

Discuss a direction