What clients say about working with us
Honest accounts from teams across the UAE who have used our services. We include what worked and what clients found difficult, because that is more useful than a curated highlight reel.
← Back to HomeWhat clients have told us
We used the computer vision tooling for incoming shipment checks. What I valued most was the scope conversation at the start — they were direct about what the tool would catch and what it probably would not. That set realistic expectations for the team. The written limit document they provided has been referenced more than I expected.
The architecture review was two hours well spent. We were about eight weeks into planning an AI feature for our internal platform and had accumulated some decisions we had not examined carefully. The written follow-up note flagged two specific assumptions we needed to revisit. One of them would have caused a significant rework later.
We have a large internal policy document library that people constantly search inefficiently. The conversational interface they built has visible citations, which matters to us — we cannot rely on answers that we cannot trace back to source documents. The quarterly reviews have been useful. They look at actual usage data, not just general impressions.
Nine weeks from first meeting to delivered tool. The model accuracy on the defect detection task was lower than I had hoped at 84%, but they told me that upfront and explained the conditions under which it fails. We adjusted the workflow so a person reviews the flagged items rather than accepting the output directly. That has worked.
Booked the architecture review before we committed engineering time to a planned AI feature. It was useful. They had read our architecture document carefully and asked questions that exposed a few gaps in our reasoning. The session itself was direct — not a sales pitch for more work. The follow-up note was short and pointed, which I appreciated.
Our HR team handles a large volume of policy questions from staff. The internal chat interface now handles a good portion of those queries with a citation attached. Staff can see where the answer comes from and check the source themselves. It has not replaced HR, but it has reduced the time spent on straightforward procedural questions noticeably.
Three detailed engagement accounts
Incoming cargo inspection at a freight forwarding operation
The client's inspection team was checking incoming cargo consignments for count discrepancies and visible damage, but the volume had grown faster than the team. Errors were getting through and only being caught downstream.
We built a vision-based counting and anomaly detection tool trained on images of the client's cargo types. The tool flags discrepancies and unusual conditions for human review — it does not make the decision; it prioritises the inspector's attention.
Count discrepancies detected at intake increased from 61% to 89% of cases within the first month of operation. The inspection time per consignment dropped by roughly 20%. Four downstream errors in the six weeks after deployment versus eleven in the comparable prior period.
"They were clear from the start about what 89% accuracy meant in practice. That clarity helped us design the right human review process around it." — Operations Director
Internal procedure and policy interface for a legal team
A legal team of fourteen people maintained 380 pages of internal procedure and precedent documents. Junior team members spent significant time searching for the right section, and inconsistent answers were being given to the same questions by different people.
We indexed the document set and built a retrieval interface with mandatory source citation. Every answer shows the section it draws from. We defined a clear list of query types the interface would refuse — questions requiring legal judgement were explicitly out of scope.
In the first quarter of operation, the team reported spending less time on internal procedure searches. The interface handled procedural queries with sufficient accuracy that the team stopped escalating those queries to senior counsel by the second month. Refusal rate was approximately 12% of queries, which the team found appropriate.
"The visible citations were the condition for us agreeing to deploy it. We needed to be able to verify every answer, not just trust the system." — Legal Counsel
Reviewing an in-progress AI recommendation feature
A software company was eight weeks into building an AI-powered recommendation feature for their platform. The CTO had growing concerns about one of the early architecture decisions — specifically, how training data would be collected and labelled at scale without introducing bias.
We read thirty pages of architecture documents in advance, prepared questions around the data collection concern and two additional areas we noticed, and ran a two-hour session with the CTO and lead engineer. We identified three specific points that warranted further consideration.
The client revised their data collection approach based on one of the three points before continuing the build. They estimated that correcting it later would have cost approximately four weeks of additional engineering time. The other two points were acknowledged but deferred to a later phase with documented reasoning.
"The follow-up note was three pages. Exactly the right length. It addressed three things, not fifteen. I could act on it the same day." — CTO
Contact Lulu Intelligence
Ready to talk about your project?
Send us a description of what you are working on and we will respond with whether and how we can help.
Get in Touch