Skip to content
Fredrik Bratten

Applied AI - Automation - Reliable agent systems

Building AI systems that can be inspected, governed and used in practice

I combine applied AI with more than 20 years in enterprise IT, automation, operations and cybersecurity. My work covers retrieval systems, agent orchestration, user-facing AI products, provenance, evaluation and controlled execution.

Explore proof packages Explore capability inventory GitHub profile

Start here

Two complementary ways to review the work

The proof packages show working systems and implementation evidence. The capability inventory shows the reusable control models, reasoning structures, context mechanisms and governance methods behind them.

Evidence route

Proof packages

Review current evidence for retrieval, agent safety, product engineering, provenance and runtime governance.

Implemented systems

Open the proof package index ->

Capability route

Capability inventory

Review reusable methods for controlled execution, reasoning under uncertainty, semantic orientation, context acquisition, capability reuse and bounded activation.

Methods + protocols

Open the capability inventory ->

Source-pinned proof packages

Four current flagships

Each package separates implementation evidence, public demonstration, source pins, limitations and non-claims.

Retrieval and graphs

Intelligence Engine

Hybrid lexical, semantic and graph retrieval with schema-driven domains, REST, MCP and an interactive web UI.

Explore ->

Human-in-the-loop agents

MADS

A controlled development environment where agent proposals become reviewable ChangeSets rather than uncontrolled host actions.

Explore ->

AI product engineering

Adaptivearts.ai

Astro, React and Supabase product architecture with editorial workflows and a server-side AI provider boundary.

Explore ->

Runtime governance

Gate Monitor

Deterministic policy for long-running agent sessions using evidence, thresholds, findings and explicit decisions.

Explore ->
Vertex visual systems proof Worktrace provenance proof Broker Lane Sandbox SPINE source-pinned proof Orchestrator evidence route

Capability inventory

The systems behind the systems

The inventory exposes reusable capabilities for controlling execution, reasoning under uncertainty, keeping context trustworthy and discovering or activating existing capability without unnecessary duplication.

Browse the capability inventory
Control and executionBound work, decisions, instructions and risky actions. Reasoning and interpretationWork with claims, uncertainty, alternatives and independent baselines. Orientation, semantics and contextResolve identity, provenance, relationships and relevant context. Capability discovery and activationFind, reconcile and reuse capability before building more. Solution-space map Explore how the 18 published capabilities combine around problem shape, lifecycle position and operating scale, including an interactive visual atlas of recurring solution patterns. Public URL: /methods/solutions/

Earlier showcases and experiments

Broader project archive

These examples remain available as historical and supporting demonstrations. Current source-pinned proof packages are listed above.

8me Orchestration

Progressive loop-orchestration labs and persistent AI workflows.

SPINE Showcase

Multi-agent orchestration, context engineering and execution surfaces.

Adaptive MCP Orchestrator

Capability routing, fallback and observability reference implementation.

arbiter

MCP protocol validation, quality checks and remediation packs.

Security Audit MCP

Security scanning and isolated analysis patterns.

agentspool

Inter-agent messaging and delivery semantics.

switchcore

MCP discovery, inventory and bounded workflow recommendations.

vigil

Durable follow-up checks with retry and expiry.

spawn

Pattern-driven MCP project generation.

Music Video Creator

Beat-aware rendering and multimodal output experiments.

AI-Human Admin Dashboard

Human and machine-readable project coordination views.

From Blueprint to Application

Book and interactive demonstrations for structured AI delivery.

How I work

Evidence before confidence

I prefer systems where implementation claims can be inspected and risky actions remain bounded.

  • - Pin source and implementation receipts.
  • - Separate tests, runtime evidence and documentation claims.
  • - Replace private operational data with synthetic public demonstrations.
  • - Keep limitations and non-claims beside strengths.
  • - Route risky actions through policy or human approval.
  • - Preserve earlier evidence when systems evolve.