A retrieval-augmented agent that answers customer product questions from the company's own data, with citations back to the source so answers can be checked instead of taken on faith. We owned the whole pipeline: ingestion, chunking, embeddings, retrieval tuning, and the evaluation harness that caught bad answers before users saw them.
A tool-using agent that pulls live shipment and inventory status straight from internal APIs and answers in chat, so staff never learn a new dashboard. We built the function-calling layer, the API integrations, and the fallback handling for when an upstream system goes slow or stops responding.
One agent, two jobs: help existing customers and convert new ones. It reads intent on each message, routes accordingly, and escalates to a human at the moment a conversation needs one. We wired in CRM reads and writes, the handoff logic, and conversation state that survives across sessions.
A voice-and-messaging agent that works an outbound list end to end. It qualifies leads, handles the usual objections, and books the good ones straight onto the calendar. We built the call flows, the scheduling integration, and the follow-up sequence so nobody on the team had to chase no-shows by hand.
QA for an AI quiz builder where students upload notes and the app generates practice quizzes. We tested the full path from upload to generation to scoring, then layered on regression and cross-browser coverage.
QA for a fintech web app handling accounts and payments, the kind of product where a rounding error or a broken auth flow isn't cosmetic. We tested the money math, the form validation, and the security-sensitive paths, then built out regression coverage and leaned hard on the edge cases around payments and login.
QA for an Android game across a matrix of real devices rather than emulators. We covered gameplay, performance, and stability on actual hardware, worked through Android's device and OS fragmentation, and reproduced and triaged crashes coming in from the field.
QA for a Windows app that locks and protects files, so the tests that mattered most were the ones where data could go missing. We verified locking, unlocking, and recovery across Windows versions, checked compatibility and data integrity, and ran the full release-readiness pass before each ship.
Ran delivery end to end in Jira: backlog grooming, sprint planning, bug triage, and release sign-off. The real job was keeping everyone honest about what was shipping and when, with clear status from first commit to launch.