Working AI Product Prototype | Enterprise Support | Agentic AI | Deterministic Execution
ResolveAI is a customer-facing AI resolution agent I built to explore a specific product question: Can an AI system move beyond giving helpful answers and safely complete a customer's task?
A walkthrough of ResolveAI investigating customer problems, recommending grounded resolutions, and executing account changes only after explicit approval.
Includes a multi-action cancellation and refund, with independent verification of both outcomes.
Based on actual acceptance tests using synthetic customer data.
ResolveAI_Portfolio_Walkthrough.mp4
Detailed implementation and test evidence are available on request; the source repository remains private.
Scope: Independently developed portfolio prototype using synthetic customer and account data. Verified scenarios demonstrate working flows, not production-scale reliability or measured business impact.
Support experiences frequently require customers to select issue categories, repeat account information, and navigate separate workflows for billing, activation, cancellation, and refunds. A chatbot may explain the issue but still leave the customer to complete the task themselves.
I designed ResolveAI around a more meaningful outcome: resolve the customer's actual problem with appropriate evidence, controlled account changes, and verified completion.
A customer simply describes a problem in natural language. ResolveAI investigates account facts and relevant policy, asks only when clarification is needed, recommends a next step, and seeks explicit approval before any consequential change. It then executes through deterministic code, independently verifies the result, and resolves or escalates the case.
Example: “I think I've been charged twice this month. Can you check?” The agent identified two $79.99 charges occurring four seconds apart, checked refund history, applied the applicable duplicate-charge policy, proposed one refund, requested confirmation, and verified the completed refund.
ResolveAI is designed around a simple but important operating model: AI investigates and recommends, while deterministic systems authorize, execute and verify consequential actions.
This separation allows the product to stay natural-language-first for the customer while maintaining clear approval, policy and verification controls.