Independent AI systems studio

AI systems that ship and keep working.

Architecture, integration, and evaluation for teams building with LLMs and agents. Explicit where possible. Learned where necessary.

  • System architecture
  • Agent integration
  • Evals & reliability
decision_boundary.v26 live

Symbolic

schema trace eval contract
S×S

Subsymbolic

agent model latent
Remote / global Select availability · Q3 2026

Working model

The seam is the system.

Explicit where possible. Learned where necessary. That's how AI features stay shippable.

Most AI features fail the same way: a promising demo gets brittle the moment real users touch it. Too much is left to the model that should have been engineered—or too much is engineered that the model handles better.

SXS helps teams draw that line well: the structure around the model, the evals that prove it works, and the integration patterns that keep it working as models and requirements change.

Structure

  • Architecture and schema design
  • MCP contracts and agent tool specs
  • Event-driven systems and observability

Intelligence

  • LLM, RAG, and agent integration
  • Evaluation and reliability frameworks
  • Model-in-the-loop product design

“Build what you can specify. Learn what you can't.”

Engagements

Get to a sound decision faster.

Focused technical sessions for a live decision, an AI feature moving toward production, or a system that needs a deeper audit.

QUICK CONSULT 01
$249 15 min
  • Unblock one specific technical decision
  • Direct answer on architecture or tooling
  • Written action items you can hand to your team
  • Curated resources — no homework required
Discuss this engagement
COMPREHENSIVE 03
$699 60 min
  • Full architecture audit with written findings
  • Eval strategy so you can prove it works
  • Tooling recommendations matched to your team
  • Enablement plan to make your team self-sufficient
  • 30-day async support while you implement
  • Code review included
Discuss this engagement

Remote by default. A short written brief follows every session.

Need an embedded engagement? Start with the system and the constraint ↗

Selected work · 2026

Current systems, not case-study theater.

Original products, active integrations, and maintained open-source work. The newest build appears first.

02 Thinking Sage · Agent capability BUILDING

2026

ADR POWER

A portable agent capability for creating, reviewing, cross-referencing, and updating architecture decision records inside a codebase.

03 Thinking Sage · Platform BUILDING

2026

CONTEXT BAZAAR · KANON

A canonical knowledge-artifact system for AI coding assistants. Author a skill, rule, workflow, or agent once; Kanon validates and compiles it for Codex, Claude Code, Kiro, Copilot, Cursor, Windsurf, Cline, and Q Developer.

04 Product variant / maintained fork LIVE

2026

WORLD MONITOR · SXS

An SXS product variant of the real-time global intelligence dashboard: AI-synthesized briefs, 45 map layers, five focused views, and native desktop builds.

05 Thinking Sage · Research product BUILDING

2026

RHIZA

An AI-assisted etymology explorer that traces the Greek roots of English words and turns their relationships into interactive, clustered graphs.

06 Thinking Sage · Creative tooling MAINTAINED

2026

BYRON POWERS

Eight literary and publishing workflows for novelists, technical authors, agents, and proofreaders—distributed as reusable AI capabilities.

Browse the full repository history

Field kit

A stack is a set of tradeoffs.

These are the tools in current use—not a wall of every logo encountered. The system, team, and operating constraints make the final call.

01

AI SYSTEMS

Models are components, not architecture.

OpenAI CodexClaudeKiroMCPOllamaRAG + evals
02

LANGUAGES

Chosen for the shape of the system.

TypeScriptRustPythonGo
03

INTERFACES

Web, desktop, terminal, and data-rich UI.

SvelteKitViteTauriRatatuiD3.js
04

PLATFORM

Deployment that the operating team can own.

AWS + BedrockCloudflareVercelDockerGitHub Actions
05

DATA

Structured for traceability and change.

PostgreSQLNeo4jRedisProtobufVector search
06

PRACTICE

The controls that keep a demo operable.

ArchitectureADRsObservabilitySecurityCI/CD

Next step

Bring the constraint, not the pitch deck.

A useful first note is short: what you're building, where it breaks, and the decision that needs to be made.

A strong first message includes

  1. 01The system or feature in one sentence.
  2. 02The current technical constraint.
  3. 03The decision or outcome you need.

Availability Select projects

FormatRemote / global

ResponseWithin 1 business day

Open a direct thread

[email protected]

No intake portal. No message that pretends it was sent.