Scientific curiosity. Engineering discipline.
Explore broadly, then narrow the system until its behavior can be explained, measured, and maintained.
01AI ENGINEERING / AGENTIC SYSTEMS
Harnessing intelligenceinto working systems.
I build the system around the model—context, memory, tools, evaluation loops, and safety boundaries. Capability matters when it becomes reliable behavior.
HARNESS://ORCHESTRATOR● LIVE
RCHUMAN
CONTROL
CONTROL
MODELreason
CONTEXTremember
TOOLSact
EVALverify
$ compose --intent useful --mode reliable_
02SELECTED SYSTEMS
Proof lives in the work.
A small selection of systems built to make model behavior more useful, legible, and controlled.
SYS.010 RUNTIME DEPS
Agent Tool Safety
A deterministic Python policy gate that turns agent tool calls into auditable allow, deny, or confirm decisions—before execution.
PYTHONAI SAFETYPOLICY
INSPECT REPOSITORY ↗
SYS.02A2A × MCP
A2A Healthcare Systems
An end-to-end agent network: orchestrator, policy, research, and provider agents communicating across A2A and MCP.
A2AMCPMULTI-AGENT
INSPECT REPOSITORY ↗
SYS.03SOURCE AWARE
RAG with Frontier Models
A production-oriented retrieval API with FastAPI, PostgreSQL, pgvector, document pipelines, health checks, and cited answers.
RAGFASTAPIPGVECTOR
INSPECT REPOSITORY ↗
SYS.04DOMAIN ADAPTED
Clinical NLP Systems
A clinical language pipeline that combines specialist models, builds domain-ready training data, and adapts information extraction for medical text.
CLINICAL NLPDOMAIN DATAHEALTHCARE AI
INSPECT REPOSITORY ↗
03THE HARNESS
The model is only one component.
Reliable agentic software is interface design: what the model can see, what it can do, how its work is checked, and where people remain in control.
PIPELINE://USEFUL_BEHAVIORTRACE_ID RC-042
01MODEL
02CONTEXT
03TOOLS
04FEEDBACK
POLICY GATEALLOW / CONFIRM / DENY
EVALUATIONEVIDENCE / TRACE / VERIFY
HUMAN CONTROLVISIBLE / REVERSIBLE / OWNED
OUTPUT STABLE · USEFUL BEHAVIOR
04OPERATING PRINCIPLES
Quiet systems. Clear boundaries.
Open knowledge compounds.
Useful ideas should leave the notebook. Open source turns personal learning into shared infrastructure.
Humans remain in the loop.
Safety, privacy, and trust are not finishing layers. They shape the architecture from the first interface.
05OPEN SOURCE
Ideas become infrastructure when shared.
Open work across research, education, and production systems. Built to be studied, challenged, and improved together.
OPEN-SOURCE WORK ↗WHERE IT SHIPS
Giraffe Analytics
A focused technology studio turning AI systems into pragmatic, reliable production work.
VISIT COMPANY ↗