Feedback From People Who Worked With Us

The audit was more thorough than I expected. They identified three intent categories where our bot was failing consistently — things we had not noticed because we were looking at aggregate metrics rather than specific conversation flows. The improvement plan was practical, not a wishlist.

The vision pipeline reduced our manual inspection time significantly. There were a few rounds of refinement needed on certain defect types because our sample images were not as varied as they needed to be — the team explained this clearly and we worked through it. The final system met the threshold we agreed at the start.

Good communication throughout the project. We had daily activity updates and one formal check-in each week. I appreciated that they flagged a data quality issue in week two rather than discovering it at handover. The personalisation engine has been running in production for two months with no major issues.

We engaged mynexadra for a chatbot audit after seeing declining containment rates for six months. The root cause turned out to be in training data labelling, not the model itself. That insight alone saved us from a costly rebuild. The written report was detailed enough for our in-house team to act on without further input.

The recommendation system delivered a measurable improvement in click-through on our homepage placements. The two-week optimisation window after launch was particularly useful — the team came back with adjustments based on real traffic that the offline testing had not anticipated. Documentation was thorough.

What stood out was how specifically they scoped the vision pipeline to our actual data. We do label inspection for packaging and the variation in lighting conditions was a real challenge. They proposed a data augmentation strategy that addressed it during training rather than at inference time. That made the deployed model much more reliable.