60+
Engagements completed
4.7
Avg. satisfaction score
6
Years in Singapore
94%
Clients who re-engaged
Experiences from recent engagements
Wei Cheng Lim
Head of AI Risk, Regional Bank · SG
We had models in production that our team was essentially flying blind on. Neuralith set up a full monitoring layer in about three weeks — dashboards, drift detection, alert routing. What I valued most was the runbook. When something flags at 11pm, the operations team now knows exactly what to do without calling me.
February 2026 · AI Observability Engagement
Priya Krishnan
VP Data Science, Logistics Firm · SG
We needed to model supplier risk across a multi-tier network — standard regression simply could not capture the relational structure. Neuralith was careful to validate whether GNNs were the right fit before committing to the architecture. That honesty at the start meant we did not waste weeks on the wrong approach. The solution has been in production for five months now.
January 2026 · Graph Neural Network Engagement
Daniel Toh
CTO, Fintech Startup · SG
We were evaluating three AI/ML platforms and getting contradictory advice from every vendor. Neuralith ran a structured evaluation — real POC testing against our actual data — and gave us a clear recommendation with the reasoning behind it. The engagement ran slightly over the original estimate due to our own delays, but the deliverables were exactly as described.
December 2025 · Enterprise Platform Selection
Nurul Syahirah
Data Engineering Lead, Healthcare Group · SG
Given the sensitivity of data in our environment, the way Neuralith handled data access was important to us. They scoped the minimum data needed, signed a proper NDA immediately, and the PDPA alignment they built into the monitoring design was thorough. Our DPO reviewed the architecture and had no concerns — which is not a common outcome.
January 2026 · AI Observability Engagement
Aaron Ong
Principal Engineer, E-Commerce Platform · SG
The knowledge transfer aspect was something I was sceptical about going in — most engagements promise this and deliver slide decks. Neuralith actually ran structured sessions with our engineers, reviewed their work as part of the handover, and left documentation we have used consistently since. Three months after the engagement closed, we extended the monitoring ourselves without needing to call them.
November 2025 · AI Observability Engagement
Josephine Lau
Head of Analytics, Insurance Group · SG
We were moving off a legacy MLflow setup that had accumulated years of technical debt. Neuralith assessed our current state, recommended a migration path to a modern platform, and handled the migration planning alongside the deployment. The scope was clear from day one and they flagged one complexity early that could have become a serious delay if left until later.
February 2026 · Enterprise Platform Setup
Engagement stories in more detail
Production model monitoring for a regional trade finance platform
Challenge
A Singapore-based trade finance platform had three ML models in production — credit risk scoring, fraud signal detection, and document classification — with no systematic monitoring. Issues were discovered through downstream complaints, not direct observation. The engineering team lacked the tooling and methodology to change this.
Solution
Neuralith implemented a layered monitoring system covering all three models: per-model performance metrics, data drift detection with configurable thresholds, prediction confidence distribution tracking, and alert routing to the on-call rotation. A monitoring playbook was developed for each model covering the three most common failure modes and response steps.
Results
Within six weeks of handover, the system surfaced a drift event in the document classification model that would previously have gone undetected for months. The operations team responded using the runbook without engineering escalation. The engagement ran four weeks and came in within budget.
4-week engagement · SGD 490
Graph-based fraud detection for a payment network operator
Challenge
A payment network operator was seeing coordinated fraud patterns that their existing rule-based and tabular ML systems consistently missed. The fraud was network-structured — coordinated rings rather than isolated transactions — and the relational signal was not captured in any of their current features.
Solution
After a graph suitability assessment confirmed the approach was warranted, Neuralith modelled the transaction network as a heterogeneous graph, selected a graph attention network architecture appropriate to the edge types, and built the training pipeline with iterative evaluation. Integration with the existing data infrastructure was included in the scope.
Results
The GNN-based detector achieved a 31% improvement in recall on ring fraud cases versus the best-performing tabular model, at a comparable precision level. The engagement ran eight weeks with two evaluation checkpoints. The client's data science team was able to retrain and evaluate the model independently at the close of the engagement.
8-week engagement · SGD 1,750
AI platform selection and migration for a mid-sized insurer
Challenge
A mid-sized insurer had been operating on a self-managed Jupyter-and-MLflow stack that had outgrown the team's capacity to maintain it. Experiments were poorly tracked, model versioning was inconsistent, and onboarding new data scientists took weeks. They needed to move but were receiving conflicting advice from vendors.
Solution
Neuralith conducted a requirements analysis with four stakeholder groups, produced a structured evaluation of three shortlisted platforms against the organisation's specific constraints, and ran a four-week POC phase with the top two candidates using the insurer's actual workflows. The recommended platform was then deployed with migration of existing experiment history included.
Results
The deployment was completed in twelve weeks. The insurer's data science lead reported that new data scientist onboarding time dropped from roughly three weeks to four days after the platform change. The architecture decision record produced during the engagement was used internally six months later when evaluating a related tooling question.
12-week engagement · SGD 2,480
Ready to discuss your situation?
We respond to all enquiries within one business day. Initial conversations are confidential and there is no obligation to proceed.
Robinson 77, Singapore 068896
Mon–Fri: 9:00–18:00 SGT
Sat: 10:00–14:00
Have an AI challenge worth
looking at carefully?
Tell us what you are working on. We will tell you honestly whether we think we can help.
Request a Consultation