AI engineer

Anton Wiklund

Agents, RAG, evals, and products.

Work

Shipped AI work.

01Teton.ai

I built AI agents. For care teams.

I was the first AI agent engineer hired at Teton.ai in Copenhagen. I worked on Samwise, a complete agent harness for care teams.

Samwise runs in a sandbox, has its own DSL, creates PowerPoint, PDF, and DOCX files, and can use sites, policies, documents, learnings, and other context. I also worked on specialized agents, including a 3D-room agent that fills out uploaded fall-incident forms. Adoption was 10x higher than when I started.

AI agentsSamwiseSandbox DSLEvalsDocuments3D room data10x adoption

View Teton.ai

02EasyFindAI

A support bot. Built from legal RAG.

EasyFindAI came from DocumentFlow. I reused the legal-grade RAG work and turned it into a support bot for websites.

The value was accuracy and setup speed: evals for answer quality, four clicks to create a bot, then one script line pasted into a website.

Legal-grade RAGEvals4-click setupOne-line embedCrawlerRAGNeo4jPinecone

View EasyFindAI

03DocumentFlow

Legal documents. Easy to search.

I built DocumentFlow as a desktop app for legal work, with a native Tauri client and a Python/FastAPI backend on AWS.

The core is a strong RAG system with Pinecone and a Neo4j knowledge graph. When a document changes, it updates only the affected chunks and graph links instead of rebuilding everything.

TauriReactPythonFastAPINeo4jPineconeGraph updatesOAuth2 + MFAMongoDB

View DocumentFlow

04SmartPrompt

Prompts you can test. And reuse.

I built SmartPrompt after needing a better prompt workflow inside DocumentFlow. It gives users private workspaces for prompts.

It supports variables, live token counts, OpenAI test runs, saved versions, sharing, and one-click LangChain exports.

Next.jsTypeScriptClerkDrizzlePostgreSQLtiktokenOpenAILangChain export

View SmartPrompt

05RAGCheck

Check AI search. See what breaks.

I built RAGCheck to test retrieval systems automatically. It generates questions from documents and runs them across multiple LLMs.

The dashboard shows pass/fail reasons, model comparisons, and weak spots like missing context, bad sources, or answers that sound right but are wrong.

PythonMulti-LLMAuto Q&ABatch testsStreamlitPlotly

View RAGCheck

About

I build AI that ships.

I work across agents, retrieval, evals, and product code. I care about systems that users can trust and teams can maintain.

Education

Grit:lab AI track

Foundation

Aland Vocational IT

Skills Education