Skip to content

Experiments Registry

What we are testing

Each experiment below tests a specific hypothesis about how AI systems can be built, deployed, and operated. Status tracked from hypothesis to verified finding.

Verified

Agent Memory Persistence

Persistent cross-session memory significantly reduces agent repetition and improves task completion for multi-turn autonomous agent workflows.

Findings: Memory persistence reduced average task completion time by 40% and eliminated redundant context re-establishment. Key in...

AgentsMemoryInfrastructure
Verified

Verification Gate Pattern

Structured verification gates at defined checkpoints in autonomous agent workflows will catch hallucinations and drift before they compound.

Findings: Verification gates caught 78% of critical errors before they reached production. The key design principle: gates must be...

AgentsQualityArchitecture
Experimental

Android-Native Coding Agent

A mobile-first coding agent can handle repository-level operations (branch, commit, review) from an Android device while maintaining the same quality discipline as desktop workflows.

Findings: Repository browsing and review work well on mobile. Full-scale code generation benefits from a hybrid approach (draft on...

MobileDeveloper toolsAndroid
Verified

Multi-Format Document Understanding

A single pipeline can handle PDFs, images, spreadsheets, and handwritten documents with comparable accuracy to format-specific solutions.

Findings: Achieved 92% extraction accuracy across formats. The unified schema was the critical design decision — format-specific p...

DocumentsAIProcessing
In Development

Multi-Agent Orchestration Pattern

A coordinator-worker pattern with shared memory and verification gates will outperform single-agent architectures for complex multi-domain tasks.

Findings: Early results: 3-agent configuration outperforms single-agent on multi-domain tasks by 35%. Main challenge is inter-agen...

AgentsOrchestrationArchitecture
Verified

Construction Field Data System

Real-time field data capture with live mapping will reduce construction project coordination overhead by at least 30% compared to paper-based or spreadsheet workflows.

Findings: Coordination overhead reduced by an estimated 45% in pilot deployment. The live map was the highest-impact feature — spa...

Field opsMapsConstruction