A portfolio with visible maturity boundaries.
This index separates what has been built and evaluated from research underway and directions that remain architectural exploration. It is designed to make the state of the work legible.
Reusable patterns for retrieval, memory and evaluation.
These projects focus on the infrastructure and control patterns that make AI systems more inspectable.
ARMIE Retrieval Platform
Evaluation-driven retrieval systems engineering across dense, sparse and hybrid retrieval, deterministic constraints, governed natural-language interpretation, observability and reproducible validation.
v0.5.1 · Governed NL → confirmed RetrievalContract → deterministic Elasticsearch execution.
Project detail ↗Sovereign Companion Runtime
A local-first AI companion runtime designed so identity, memory, relationship state and continuity remain portable across replaceable language models.
Founder Workbench, persistent Runtime state, local-model execution, provenance and systematic generalization evaluation.
Project detail ↗Domain execution with evidence and bounded claims.
Vertical projects apply the same discipline to a specific information environment without implying a unified production platform.
Construction Intelligence Workbench
Auditable BIM and multimodal AI for typed planning, deterministic domain execution, evidence, independent verification and an inspectable audit trail.
Synthetic BIM, drawing and viewer workflows in one bounded workspace.
Project detail ↗Healthcare AI Platform · AWS
An AWS-native safety and evaluation workbench with authenticated execution, dual async orchestration, persistent evidence and a live Bedrock model under an independent safety boundary.
Cognito, Step Functions/SQS, DynamoDB/S3, live Bedrock execution and capability regression evaluation.
Project detail ↗Healthcare AI Platform · Azure
A synthetic healthcare AI reference system combining a Python reasoning runtime with authenticated Azure execution, Durable Functions orchestration, persistent state/evidence, independent safety validation and a React Workbench.
Azure Dev runtime, identity, durable orchestration, Cosmos/ADLS persistence and Workbench integration.
Project detail ↗Architecture questions being developed, not released products.
These directions have defined intentions and constraints; their practical viability remains an open research question.
Personal AI Superbrain
A direction for traceable personal long-term memory retrieval across events, time, people, documents and other evidence.
Memory representation, provenance and retrieval boundaries.
Project detail ↗