Multi-Agent Systems
Identity, scoped context, durable reviewed memory, project rooms, messaging, and conflict-aware coordination for simultaneously operating external agents.
Review multi-agent architecture
Machine Intelligence Engineer
Software Architect
Multi-Agent Systems
Michael Kappel — Machine Intelligence Engineer / Software Architect
I build multi-agent systems, persistent AI memory, and LLM infrastructure.
I architect machine-intelligence software for agent coordination, durable context, local model/runtime tooling, governed workflows, and enterprise AI integration. More than two decades of .NET, Azure, SQL, TypeScript, testing, and production architecture provide the engineering foundation.
Review Multi-Agent Systems
Open Multi-Agent Memory
Companion docs
Intelligence724 capability
Resume
Email Mike
20+ years enterprise delivery
MATM multi-agent coordination and reviewed memory
32 public NuGet packages
21K+ cumulative NuGet downloads
Best fit for
AI systems roles that need production engineering discipline.
Multi-agent and LLM infrastructure first; enterprise architecture evidence remains one click away.
Machine Intelligence / AI Systems Architecture Coordination, memory, local runtime components, retrieval, validation, recovery, and enterprise integration.
Multi-Agent and AI Memory Platforms Registered identity, scoped context, reviewed memory, rooms, messages, edit claims, leases, and human verification.
AI Developer Tools / LLM Infrastructure .NET runtime and package components with explicit capability boundaries, diagnostics, and a verified managed-CPU path.
Choose a review path
Start with the strongest public proof.
The deep evidence remains available, but first-time visitors can reach machine-intelligence proof before the broader career archive.
Front-door map
Five ways to verify the positioning.
Each route answers a distinct reviewer question and links directly to public evidence.
01 Machine Intelligence Professional identity, architecture pillars, target roles, and evidence boundaries.
02 Multi-Agent Memory Problem, architecture, validation, public source, and exact non-claims.
03 LLM Infrastructure 32 public packages, 21K+ downloads, runtime families, and capability limits.
04 Intelligence724 Applied machine-intelligence capability for code intelligence, AI documentation, modernization, retrieval, and bounded automation.
05 Resume Machine-intelligence summary with the full enterprise career foundation.
About & skills
Machine-intelligence systems backed by enterprise engineering.
Michael Kappel is a Machine Intelligence Engineer and Software Architect specializing in multi-agent coordination, persistent AI memory, local LLM infrastructure, governed workflows, and enterprise integration. The supporting foundation spans more than two decades of .NET, Azure, SQL, TypeScript, testing, modernization, and technical mentoring.
Persistent AI Memory
Typed local .uai continuity plus authenticated shared memory, lifecycle governance, recovery, and human verification.
Open AI memory architectureLLM Infrastructure
Local GGUF/LLaMA-oriented .NET runtime components, model APIs, retrieval, package engineering, and observable backend boundaries.
Explore LLM infrastructureEnterprise AI Integration
Governed model workflows, semantic retrieval, structured outputs, recovery, and human review integrated with business-critical software.
Open machine intelligence overviewIntelligence724 applied AI capability
Process-first AI modernization for existing systems: code intelligence, AI documentation, governed knowledge/retrieval, legacy modernization, legal/intake workflow patterns, and bounded automation.
Visit Intelligence724Enterprise Engineering Foundation
20+ years across Microsoft Azure, .NET, SQL Server, TypeScript/Angular, testing, modernization, and production architecture.
Review enterprise experiencePhotography / visual creativity
A separate image-forward portfolio showing composition, public media archives, visual systems, and creative range.
Open photography portfolio
What I build
Machine-intelligence infrastructure that can be inspected, governed, and recovered.
Each capability is tied to public source, registry evidence, user-confirmed work, or an explicit claim boundary.
Multi-agent coordination
Give external agents registered identity, governed scope, shared project context, routed messages, and cooperative conflict controls.
View multi-agent architecturePersistent AI memory
Separate first-load local continuity from authenticated shared memory, review lifecycle, recovery, and evidence-backed readback.
View AI memory architectureLLM/runtime infrastructure
Build contracts and components around local models, APIs, retrieval, structured output, diagnostics, observability, and fail-closed capability boundaries.
View LLM infrastructureMachine-intelligence architecture
From model capability to governed enterprise system.
The architecture keeps agent execution, coordination, memory, and enterprise application responsibilities explicit.
Models + external agents local runtimes, model APIs, specialized agent responsibilities
Coordination + memory identity, scopes, rooms, messages, reviewed context, claims, leases
Enterprise application layer .NET APIs, Azure, SQL, TypeScript, retrieval, validation, recovery
Human governance source boundaries, review, observability, escalation, auditable evidence
Text alternative: models and external agents sit above a coordination and memory layer. Enterprise applications integrate those capabilities through .NET, Azure, SQL, TypeScript, retrieval, validation, and recovery. Human governance provides review, escalation, observability, and source-bounded evidence.
Selected systems & case studies
Machine-intelligence proof before the broader project archive.
The default set is intentionally short; the full evidence surfaces stay available when a deeper technical review is needed.
Public Source Available Reference Implementation
Multi-Agent Memory Coordination System
A deployable private-intranet, source-available reference implementation for coordinating external AI agents through registered identities and scoped grants, persistent reviewed memory, meeting and current-message routing, acknowledgements, and hash/CAS/lease controls for simultaneous local .uai edits.
Validation Review the public source and companion documentation, run the repository validation suite, and trace identity, scope, memory, message, acknowledgement, hash/CAS, claim, lease, and persistence contracts while checking every explicit non-claim.
.NET 9 C# HTTP APIs JSON SQLite
/machine-intelligence/ /multi-agent-systems/ Resume Backed Professional Pattern
Enterprise Modernization and Parity Validation
A resume-backed case study for modernizing legacy Web Forms, Classic ASP, and SQL-heavy systems while preserving business behavior through comparison screens, generated scenarios, and test coverage.
Validation Use behavioral comparison screens, generated scenario tests, automated unit/integration testing, and source review before treating new code as behavior-compatible.
ASP.NET Core C# SQL Server T-SQL stored procedures
/dotnet-sql-modernization/ /evidence-map/ Public Platform
UAIX AI Memory Package Wizard
A public-platform case study for AI memory, file handoff, project handoff, and compact AI handoff state mapped to long-term Architecture Notes research.
Validation Validate JSON exports, mirror alignment, package read order, and generated-needs-review status for safety anchors before release packaging.
UAIX AI handoff AI Memory Project Handoff File Handoff
/ /case-studies/ Experience & engineering foundation
The production discipline behind the AI systems work.
Use the career and experience routes for formal titles, employment history, and the complete Microsoft Azure evidence sequence.
20+ years enterprise software
.NET, SQL Server, APIs, reporting, line-of-business workflows, modernization, and production support.
Experience routeAzure in every role since 2010
Role-specific App Service, Azure SQL, Blob Storage, VMs, DevOps, Foundry, Search, and earlier Windows Azure evidence.
Career overviewArchitecture and quality judgment
Business-rule translation, testing seams, API boundaries, recovery, reviewer-readable documentation, and mentoring.
Evidence mapTechnical appendix
Deeper proof stays available after the core hiring path.
These cards remain secondary to the resume and selected proof path, but keep source-backed evidence available for reviewers who need it.
Evidence lane
Multi-agent coordination and persistent memory
A source-available, deployable private-intranet reference implementation for coordinating external AI agents through identity, scope, shared context, reviewed memory, messaging, and conflict-aware work ownership.
- Public source and companion documentation expose registered agent identities, company/workspace/project scopes, memory submit/search, meeting rooms, current messages, acknowledgements, redacted receipts, and human verification.
- Hash-only file heads, compare-and-swap checks, bounded edit claims, and expiring ownership leases support cooperative simultaneous work on local .uai paths.
Evidence lane
AI Memory and handoff architecture
UAIX-style package wizard support, Project Handoff, Agent File Handoff, compact AI handoff memory, Architecture Notes, manifest exports, and receiver startup packets.
- Added a repository-local .uai package with startup, receiver brief, short-term memory, long-term memory pointer, file handoff, deployment/test report, and export manifests.
- Moved research Markdown into /docs and mapped those long-memory files from .uai/short-term-memory.uai and .uai/long-term-memory.uai.
Evidence lane
Intelligence724 applied machine-intelligence capability surface
A public process-first consulting and implementation surface for AI integration, source-code intelligence, governed retrieval, legacy modernization, bounded automation, recovery-aware workflows, and human-reviewed delivery for existing business systems.
- Intelligence724.com explains applied AI integration for existing systems, Current AI Delivery Experience, bounded agentic automation, legal operations, enterprise knowledge retrieval, and legacy-to-AI modernization at capability level.
- The relationship is professional-positioning evidence: MikeKappel.com connects Michael Kappel’s enterprise software history and Info724 / Insurance724 experience to current machine-intelligence implementation; Intelligence724 is not treated as a miscellaneous website or a replacement employer record.
Evidence lane
.NET local LLM infrastructure
A verified public NuGet ecosystem covering local LLM runtime contracts, GGUF/LLaMA-oriented components, tensors, tokenization, sampling, managed CPU and accelerator backends, AI memory, interoperability, browser/P2P tooling, governance, and observability.
- The official Michael.Kappel NuGet profile lists 32 public packages with 21K+ cumulative downloads.
- The UAIX.LmRuntime family provides a verified managed-CPU path and explicit backend registration and diagnostics; package presence alone is not presented as proof of GPU inference.
Technical appendix
Research and machine-readable files stay secondary.
Architecture Notes remain curated public documentation. Reports, search, dashboard routes, and machine-readable files remain secondary utility surfaces with explicit boundaries.
Structured review files
Structured files support tools without leading the human funnel.
Structured review files provide manifests, route indexes, JSON-LD, boundaries, and recommended reading order. They do not permit private probing or tool execution.
/llms.txt /llms-full.txt /route-qa-contract.json /api/public-route-index.json
AI memory & wiki routing
MikeKappel.com uses visible handoff paths for agents and humans.
The resume site keeps the human hiring path first while exposing reviewed AI-memory routes for agents that need source boundaries, intake status, and long-memory context.
Reviewed long memory
AIWikis MikeKappel namespace
AIWikis is the configured reviewed-memory destination for promoted MikeKappel.com setup facts. It routes agents back to the source site and does not replace source authority.
Open AIWikis memoryLLM Wiki method
LLMWikis setup strategy
LLMWikis provides the planning model for source policy, trust labels, reading order, evidence routing, and review gates before durable memory is promoted.
Open LLMWikis wizardActive intake
UAIX File Handoff setup
Root agent-file-handoff/Content/ and agent-file-handoff/Improvement/ are active intake buckets. Agents must enumerate, use, record, preserve, and remove or justify any active file.
Resume and contact
Download the two-page resume or start a direct conversation.
Email, phone, LinkedIn, GitHub, and location are shown from approved resume/contact artifacts. Use the direct links below for the fastest contact path.
Optional browser-local tool Role Matcher Kept collapsed so the homepage remains a concise reviewer dashboard.
Paste a job description, recruiter note, or technical requirement only when tool-based matching is useful. The analysis remains browser-local and the manual evidence paths stay available.
Browser-local. No pasted text is sent anywhere. Use the manual evidence map if a tool-based review is not needed.
Claim boundary
Evidence-backed portfolio, not a certification surface.
Research and handoff records are review material. This site does not claim consciousness, biological equivalence, formal standards approval, SOC 2/ISO compliance, accessibility certification, or production access to private systems.
FAQ
Common review questions.
Short answers are rendered in the HTML source and mirrored into JSON-LD where appropriate.
What is Michael Kappel's primary technical focus?
Michael Kappel is a Machine Intelligence Engineer and Software Architect focused on multi-agent systems, persistent AI memory, local LLM infrastructure, governed AI workflows, and enterprise AI integration.
How long has Michael Kappel used Microsoft Azure professionally?
The public career record documents user-confirmed Microsoft Azure use in every professional role from 2010 to the present, with role-specific services and purposes on the Experience page.
Where should a hiring reviewer start?
Start with Machine Intelligence and the Multi-Agent Memory case study, then use the two-page resume, selected projects, and experience timeline for progressively deeper evidence.