What Clients
Say About Our Work
Feedback from organisations that have engaged Sovanta for AI engineering projects in Malaysia.
← Back to Home40+
Projects Delivered
4.8
Average Rating
96%
On-Time Delivery
6+
Years Experience
What Clients Are Saying
"The data quality assessment was genuinely useful — not just a list of issues, but a clear prioritisation of what to fix before we started model development. Saved us from building on a shaky foundation."
Hafiz Tarmizi
Head of Data, Retail Platform · KL
March 2026
"We had tried two other providers for voice interface work. Sovanta was the first to actually sit down with us and map the user journey before writing any code. The end result worked properly from day one."
Lim Wei Ling
Product Manager, Fintech App · Penang
February 2026
"Our articles were not being categorised accurately and it was creating problems downstream. The classification engine Sovanta built handles multi-label cases well — something we were told by others was not feasible with our dataset size."
Nurul Khalidah
Editorial Technology Lead · KL
March 2026
"What I appreciated most was the handover process. By the end of the engagement, my team actually understood how the system worked. We were not left with a black box we had to go back to Sovanta to modify."
Raj Balakrishnan
Engineering Manager, Logistics SaaS · KL
January 2026
"The assessment report was structured clearly and the remediation plan was practical — not just a wish list of improvements. We used it to prioritise three months of internal data work before starting the classification project."
Ahmad Taufiq
Senior Data Analyst, Manufacturing · Selangor
February 2026
"Communication throughout was calm and professional. When a constraint came up mid-project that affected the original scope, they flagged it early and proposed options. No surprises at handover."
Siti Yasmin
CTO, E-commerce Platform · KL
March 2026
Project Outcomes in Detail
Case Study 01 · Content Classification
Kuala Lumpur Media Group: Automating Article Taxonomy
Challenge
A digital media team was manually tagging 300–400 articles weekly across a 60-category taxonomy. Inconsistency between editors was creating search and recommendation problems. The process consumed roughly 12 hours of editorial time per week.
Solution
Sovanta revised the taxonomy from 60 to 38 logically consistent categories, prepared labelled training data from the archive, and built a multi-label classifier that integrates with the team's CMS via API. Duration: 8 weeks.
Results
Automated tagging now handles 85% of articles without editorial review. Tagging consistency improved from approximately 70% to 91%. Editorial team redirected around 10 hours per week to other work.
"Reorganising the taxonomy before building the model made a significant difference. We had been trying to classify into categories that overlapped too much."
— Editorial Technology Lead
Case Study 02 · Data Quality Assessment
Selangor Manufacturer: Dataset Readiness Before ML Investment
Challenge
A manufacturing company was considering an ML-based quality inspection project. Leadership wanted assurance that their sensor and inspection data was fit for model training before approving a larger budget.
Solution
Sovanta audited three years of sensor data and 18 months of inspection records. The assessment identified 14 distinct data quality issues and produced a prioritised remediation plan with estimated effort for each item. Duration: 3 weeks.
Results
The client addressed the top six issues over three months and then commenced a classification project with significantly higher confidence. The assessment prevented an estimated six-figure investment in a model that would have performed poorly on the original data.
Case Study 03 · Voice Interface Design
Penang Fintech App: Adding Voice Controls to Mobile Product
Challenge
A fintech product team wanted to add voice commands for common in-app actions — balance checks, transaction initiations, and navigation — without disrupting their existing React Native codebase.
Solution
Sovanta mapped 24 voice interaction scenarios, built an NLU model covering Malay and English commands, and developed an integration layer that communicated with the app's existing API. Duration: 10 weeks.
Results
Voice command accuracy reached 88% in user testing across both language variants. The feature launched with minimal additional engineering work from the internal team, as the integration documentation was comprehensive enough to follow without Sovanta's involvement.
Credentials & Affiliations
MSC Malaysia Status
Registered technology services company under Malaysia's digital economy framework.
PDPA Compliance
Data practices aligned with Malaysia's Personal Data Protection Act 2010 throughout all client engagements.
MDEC Partner Listing
Listed as an AI services provider in the Malaysia Digital Economy Corporation partner directory.
Ready to discuss your project?
A brief conversation is usually enough to determine whether Sovanta is the right fit for your situation.
Get in Touch