ShawnDemos

Independent AI systems engineering · Original systems · Real deployments

I build the systems that make AI capable of building systems.

Persistent organizational intelligence. Multi-model orchestration. Verified software engineering. Real products running on real infrastructure.

Explore live demos Review source and experiments Software Factory Coming next
ONE LOOP

Context in, verified software out, and every mistake comes back as a lesson.

 Knowledge base Memstalker · MCP Models Claude · GPT · Gemini Agents Daniel · Agent Swarm Harness & evals Gates · reviews Applications Four Presient demos Lessons Mistake ledger → KB
  1. Knowledge baseMemstalker · MCP
  2. ModelsClaude · GPT · Gemini
  3. AgentsDaniel · Agent Swarm
  4. Harness & evalsGates · reviews
  5. ApplicationsFour Presient demos
  6. LessonsMistake ledger → KB
Context in, verified software out, and every mistake comes back as a lesson.

Presient Labs demos

Bounded AI agents in realistic products

Two run live right now and all four have their source on GitHub. In each one, code decides what the agent may see and do, and every demo runs on synthetic data.

Built as Presient Labs demos, not client deliveries · github.com/Presient-Labs

The chat widget on a sample store page: a booking link, a log-in prompt for an order question, and a lead form for a bulk purchase
Open source

Website Chatbot

A Shopify store chatbot that answers only from the store's own content and says when it doesn't know.

How it works Source

Choose a path

Start with what you came for

4 projects AI engineering Software Factory · Knowledge Base Server · Daniel · Agent Swarm 5 projects Business agents CareOps · Browser Agent · Website Chatbot · Bid Selection Memo … 1 project Security & systems PwnageBox 3 projects Research & evaluation Presient Skill Evals · Presient SE Harness Pack · Memstalker Side-by-Side Demo 7 projects Products & prototypes Memstalker Landing Site · OpenClaw GUI · GenOS · MultChain Gateway …

How it connects

One system, built in layers

The projects are not separate experiments. Each is one step of the same way of building software with AI.

See the whole system →

  1. Knowledge base

    Decisions, fixes and lessons are captured once, with their source, and every agent searches the same store before it works.

    Knowledge Base Server (Memstalker)Memstalker Side-by-Side Demo

  2. Models

    The model does the reasoning and can be swapped. CareOps runs the same questions on four models to compare cost and accuracy.

    CareOpsPresient Skill Evals

  3. Agents

    Agents get roles, fail over when one runs out, and work a task graph in waves, with a watcher and test gates between waves.

    Daniel (Agent Orchestrator)Agent Swarm

  4. Harness & evals

    Stage gates, independent reviews and tests decide what counts as done. Blind evaluations check whether a change really helps.

    Presient SE Harness PackPresient Skill Evals

  5. Applications

    Bounded agents in realistic products, where server code holds the permissions. CareOps and the Browser Agent run live in Docker behind Traefik on a Linux server; the Website Chatbot and Bid Selection run locally.

    CareOpsBrowser AgentWebsite ChatbotBid Selection Memo

  6. Lessons

    In the agent swarm, review findings are recorded. Recurring ones, once approved, become warnings on the next matching task.

    Agent SwarmKnowledge Base Server (Memstalker)

The systems behind the demos

Infrastructure, research and security work

Coming next

Software Factory

A stage-gated delivery workflow where AI agents build, independent reviewers check, and a human approves. Pick a ticket for a fictional company and follow it through requirements, plan, build, tests, independent review and a human approval. Nothing is shown here until it has really run.

What it will show →

Prototype AI engineering

Agent Swarm

Keeps a team of coding agents working a task graph, with reviews and test gates between waves.

Source not public yet Case study
Research Research & evaluation

Presient SE Harness Pack

A test rig measuring whether a software-engineering harness helps an open-weights coding model.

Source not public yet Case study
Open source Security & systems

PwnageBox

A Docker lab that simulates an attack chain against a deliberately vulnerable target and logs it to Splunk.

Source on GitHub Case study

All 20 projects →

Evidence you can rerun

Numbers with the commit they came from

Each count is a full test run on the public code. Clone the repository at that commit and run the same command.

304 CareOps tests pass npm test at a0d5a47 ·
226 Browser Agent tests pass (8 paid-model tests skipped) npm test at 8f6116e ·
232 Website Chatbot tests pass npm test at ea561dd ·
Research · private repository, not rerunnable from outside 15 → 6 Runs out of 100 with edits outside the task, without and with the harness. The model fixed the bugs about 98% of the time either way. Five fixtures and one model, and a cheaper version of the same model showed no effect, so this is a narrow result. Method and limits

Dated record

Built in public, with dates you can check

Full timeline →

  1. Knowledge base server repository created
  2. Daniel multi-agent orchestrator repository created
  3. Stateless vs knowledge-base build comparison published
  4. CareOps repository created
  5. Browser agent repository created

About

One engineer, a system of agents, and evidence you can inspect

I design the products, the architecture, the acceptance tests and the infrastructure, and I direct the engineering. Coding agents do a large share of the implementation and review, inside the harness described above. Every project page says what is live, what is a prototype, and what broke and how it was fixed.

Get in touch: presientio@gmail.com

More about how I workContact