dirk wilfling/ ai portfolio
dirkwilfling.net
6 live demos — all running against Claude via Bedrock

AI/LLM Portfolio

Real enterprise problems I scoped, designed the architecture for, and delivered. Pick any demo below — they're live, not mockups.

Self-Healing Schema Agent
try it
▸ ALERT Field `marketplace_id` missing in API response
▸ STEP 1 Fetching Amazon SP-API changelog…
▸ STEP 2 Root cause: renamed → `marketplace_ids` (array)
▸ STEP 3 Generating YAML config patch…
- source: marketplace_id
+ source: marketplace_ids[0]
▸ READY Awaiting human approval → ✓ Approved

All Showcases

▸ ALERT marketplace_id missing
▸ CAUSE renamed → marketplace_ids[]
▸ PATCH ✓ awaiting approval
live

Self-Healing Schema Agent

Agentic Pipeline Repair

Turns hours of pipeline debugging into a single approval click.

Agentic AI pipelineHuman-in-the-loopSchema diff analysisConfig patch generation
Next.jsClaude via BedrockYAML configAmazon APIs
▸ QUERY "Tesla margin decline?"
▸ 3 agents cross-checking SEC data
▸ SUPREME COURT ✓ verified
live

Financial Copilot

Multi-Agent Analysis + Supreme Court

Every AI insight is cross-checked against source before you see it.

Multi-agent pipelineSupreme Court verificationRAG document searchTransparent reasoning
Next.jsClaude via BedrockRAGSEC filings
▸ companies ✓ personas ✓ scenario ✓
▸ Writing email thread…
▸ DONE 4 emails, 2 personas
live

AI Email Seed Generator

Invoice Factoring Demo Data

One click generates companies, personas, and a full email thread.

Multi-stage LLM pipelineSequential email generationPersona-driven voicesLive progress UI
Next.jsClaude via BedrockEphemeral generation
▸ "picks up kids at 3pm, no Fri"
▸ Generating schedule (§ArbZG)…
▸ All constraints met ✓
live

AI Staff Scheduler

Context-Aware Shift Planning

Understands "picks up kids at 3pm" and schedules around it.

Natural language contextAI shift generationGerman labor lawInteractive playground
Next.jsClaude via BedrockGerman labor law
▸ 847 incidents loaded
▸ Style: executive summary
▸ Report narrated ✓ 3 sections
live

AI Safety Report Generator

Construction HSE Reporting

Raw safety metrics become a professional HSE report in seconds.

LLM narration generationStyle & purpose editingTalk to the dataWhat-if scenarios
Next.jsSupabaseClaude via BedrockRecharts
▸ Prompt A → 72% accuracy
▸ Prompt B → 91% accuracy
▸ WINNER Prompt B (+19pp)
live

Prompt Lab

A/B Testing for LLM Prompts

Side-by-side prompt testing — know which one wins before you ship.

Prompt A/B testingField-level accuracyGround truth scoringEditable prompts
Next.jsClaude via BedrockDiff comparison