H2 AI Labs
All projects
Prototype / Evolving

Loose Ends

Autonomous Outcome Management

An AI agent system designed to track obligations the outside world still owes you—from refunds and returns to reimbursements and warranty outcomes.

AWS BedrockStrands Agents SDKLambdaDynamoDBTool calling

The question

What if AI tracked what the world still owed you?

01

The problem

Important outcomes disappear into inboxes, portals, and vague timelines. People must repeatedly remember, investigate, and escalate each one.

02

The idea

A persistent agent follows each obligation as a stateful case, gathers evidence, detects mismatches, and proposes the next action without taking control away from the user.

03

How it works

01Expected outcome
02Monitor
03Detect mismatch
04Investigate
05Human approval
06Act
07Verify

04 / System architecture

More than a model.

State + data

Persistent context and source evidence

AI reasoning

Interpretation and proposed action

Human control

Review, correction, and approval

Verification

Confirm the intended outcome

05 / Human + AI interaction

AI proposes

  • Monitor changing evidence
  • Identify likely mismatches
  • Investigate available paths
  • Propose an action

Humans control

  • Set the expected outcome
  • Review gathered evidence
  • Approve consequential actions
  • Confirm resolution
06

What I explored

Agent orchestrationPersistent stateTool callingHuman-in-the-loopOutcome verification
07

What worked

Modeling each obligation as explicit state made the workflow explainable and recoverable.

08

What didn't

A fully autonomous agent was the wrong goal. Ambiguous policies and consequential communications require visible human checkpoints.

09

What I would build next

Add source reliability scoring, clearer escalation policies, and richer verification across email and account portals.

Related experiments

Active BuildAI Applications

Gov Jobs Agents

AI Career Intelligence

A modular system that analyzes a résumé, explains qualification matches, ranks government opportunities, and can monitor new roles.

AI reasoningDeterministic logicPrompt versioning
Explore experiment
Concept / PrototypeHuman + AI

Creative Universe Engine

Human + AI Creative Storytelling

An AI-assisted environment helping children expand drawings and stories while preserving their authorship and decisions.

Multimodal AIImage generationPersistent characters
Explore experiment
PrototypeApplied AI

HomeForge

AI for Physical-World Projects

A safety-aware project copilot for woodworking, installation, repair, replacement, and guided DIY work.

Adaptive workflowsSafety-aware AIProject planning
Explore experiment