I turn complex business workflows into production software.

Care agents needed to establish account context, understand previous interactions, and determine the next action while speaking with a customer. I worked with stakeholders to organize the experience around that sequence.

The application brought customer details, call history, and prior incidents into the Connect workspace, with direct links into the Staff Portal for follow-up work. I shaped which information belonged in the immediate call view and how agents moved into deeper investigation.

Getting account context ready before the agent answered

My integration retrieved cached account data, compared it with the caller's supplied details, and returned the authentication result used by the automated call flow. If a mismatch or unavailable lookup prevented authentication, the caller continued to an agent without being marked authenticated, and the agent completed verification. Account context needed to reach the contact-center application in another AWS region before the caller reached an agent.

Cross-region access and a cold cache caused recurring lookup timeouts. With a broader redesign outside the project's scope, I recreated the existing Valkey lookup cache in the contact-center region and refreshed it nightly from the first region's cache. PostgreSQL remained the source of truth.

Recorded before-and-after timings for the same lookup scenarios showed the backend account-lookup request became roughly four times faster, resolving recurring timeouts. The measurement ran from request to response, including the wait for lookup data. Automated lookup used cached information until the next nightly refresh. If a refresh failed, the last successful cache remained available until a later refresh succeeded.

Connecting conversations to follow-up work

I configured Amazon Connect's built-in summarization for calls and voicemails and integrated its output into the agent application, giving agents the reason for contact, relevant discussion, and outstanding follow-up. Supporting work included voicemail handling, direct-agent dialing, and automated phone-number ingestion.

Reconstructed workflow. The contact-center region's lookup cache refreshed nightly from the first region's cache.

Those decisions shaped the wizard's branching, explanations, withholding calculations, validation, and backend behavior.

The workflow connected withdrawal choices, bank information, required authorization, and downstream processing. My broader platform work included Plaid bank verification, DocuSign signing, and ACH processing.

Earlier answers affected later steps. When a customer changed a choice, I required them to revisit the affected steps and resolve missing or invalid information before submitting. The final request needed to reflect their current selections consistently.

Resolving a signing-to-processing failure

One production issue left a distribution showing 'awaiting signature' after the customer had completed DocuSign signing.

I traced the mismatch across the portal, backend, and signing integration, corrected the association between completion events and the originating request, and repaired the status handling. I reconciled affected requests and checked the complete signing-to-processing workflow.

Reconstructed workflow. A change to an earlier answer requires affected steps to be checked before submission.

Designed and built Java Swing and Spring test-data tooling used by 50+ QA engineers to create, repair, and customize records through one application. The Swing application was a desktop interface to the Spring-backed test-data functionality.

Integrated services owned by different teams, translating user inputs into service-specific requests, coordinating multi-step calls, and handling service errors.

Worked with service owners to resolve missing capabilities and adapt integrations to available endpoints. Coordinated release testing.

Reconstructed workflow. Service names and internal systems remain private.

The old update path treated incomplete requests as replacements, allowing omitted fields to erase stored values. I implemented the Java backend and supporting frontend changes, including shared-library coordination across the Staff and Partner portals.

The correction required a consistent update contract across the API and its consumers. It eliminated the data-loss defect across several services and two portals.

Reconstructed update behavior. Omitted fields retain their existing values.

The playbook combined Argo CD deployment checks with Datadog evidence that production requests reached the new workloads, completed successfully, and showed acceptable error and latency behavior.

Release decisions were shared across the team. My responsibility was to establish confidence in the changes I worked on and surface unresolved concerns before release.

Reconstructed verification process. Deployment health and observed production behavior inform release decisions.