# Selected Impact

> A concise view of the systems, technical leadership, and learning experiences I have helped make real.


I have spent more than 15 years working across technology, architecture, and customer-facing delivery. My current focus is applied and agentic AI: turning model capability into systems with explicit controls, evaluation, and operating evidence.

## Public systems people can inspect

- **Mistral Reliability Lab:** built and maintain a Python reference with 26 credential-free tests, six deterministic evaluation cases, bounded tool execution, privacy-first telemetry, and locked-dependency auditing.
- **Runbook Relay:** built and maintain a deterministic WebMCP control room with five bounded tools, a human-only approval gate, visible receipts, and seven contract tests.
- **Public technical writing:** maintain a canonical long-form archive that connects architecture arguments to inspectable demonstrations and source code.

## Enterprise AI delivery

I work with organizations adopting AI in regulated environments. The work spans architecture, model and system evaluation, identity, policy, observability, and human approval. Public pages intentionally describe the reusable engineering patterns, not confidential customer details.

## Workshops and technical enablement

I helped run an Amazon Bedrock AgentCore Immersion Day for more than 100 participants. Within that program, I designed and delivered a separate evaluation and policy module extension for a cohort of about 30 participants. The distinction matters: the larger event demonstrates program delivery, while the smaller module reflects the material I personally built and taught.

## How I operate

- Start with the outcome, authority boundary, and failure modes.
- Turn broad capabilities into narrow contracts and observable state transitions.
- Pair architecture decisions with tests, evaluations, threat models, and reviewer paths.
- Make limitations explicit so a reference project does not masquerade as a production system.

For the implementation evidence, start with [Projects](/projects/). For the reasoning behind it, browse [Writing](/writing/). To discuss related work, [connect on LinkedIn](https://www.linkedin.com/in/andreasnissen).