Case Studies

Proof, Not Promises.
Results That Speak.

Every engagement is different. Here is a selection of real problems we solved — with the metrics to show for it.

FinTechFinancial Technology · 8 engineers · 6 months

VertexCloud: Zero-downtime migration of a $2B transaction platform to AWS

VertexCloud processed billions of dollars in transactions on a 14-year-old monolith. The platform was becoming a bottleneck — release cycles took 3 weeks and infrastructure costs were spiralling. SmallShark IT delivered a full microservices migration with live cutover.

The Challenge

Migrate 14 years of legacy .NET code to containerised Node.js microservices on Amazon EKS without missing a single transaction or incurring downtime.

Our Solution

  • Strangler-fig pattern migration — new services replacing modules one at a time, with feature flags controlling traffic routing.
  • Blue-green deployment with automated rollback triggers across 3 AWS regions.
  • Custom observability stack (Prometheus + Grafana) providing real-time business and infrastructure metrics.
99.99%
Uptime during migration
60%
Infra cost reduction
3x
Faster release cycle
$0
Data loss
HealthcareHealthcare Technology · 10 engineers · 9 months

TechBridge Health: HIPAA-compliant patient portal serving 500,000+ active users

TechBridge Health needed to replace a slow, desktop-only patient portal with a modern web application that worked on any device, passed HIPAA audits, and scaled to 500k monthly active users without performance degradation.

The Challenge

Build a greenfield HIPAA-compliant web portal with end-to-end encryption, full audit logs, SSO, and a globally distributed API averaging under 200ms response time.

Our Solution

  • Next.js 14 frontend with progressive enhancement and offline support for low-bandwidth clinic environments.
  • Node.js microservices on AWS ECS with RDS PostgreSQL — all data encrypted at rest (AES-256) and in transit (TLS 1.3).
  • Comprehensive audit logging with immutable WORM storage and automated compliance reporting.
500k+
Monthly active users
182ms
Global API p99 latency
HIPAA
Certified & audited
4.8★
App store rating
Enterprise SaaSEnterprise SaaS · 12 engineers · 8 months

NexaCorp: Real-time analytics dashboard processing 2 billion events/day

NexaCorp's data science team had valuable models but no way to surface insights to business users. Their existing BI tool was slow, un-customisable, and required SQL expertise. SmallShark IT built a purpose-built analytics platform from scratch.

The Challenge

Build a real-time analytics dashboard that could ingest 2B+ events/day, serve interactive charts in under 500ms, and be operated by non-technical business users.

Our Solution

  • Apache Kafka for event streaming + ClickHouse as the OLAP backend — delivering sub-second query responses at petabyte scale.
  • Custom React charting library with 24 chart types, drag & drop dashboards, and one-click report exports.
  • Row-level security and workspace isolation allowing enterprise clients to safely operate in a multi-tenant environment.
2B+
Events per day
320ms
Avg query response
34%
User conversion lift
40+
Enterprise clients
AI / MLArtificial Intelligence · 6 engineers · 5 months

PulseAI: Production ML pipeline with 94% prediction accuracy at scale

PulseAI had trained promising ML models in Jupyter notebooks but had no path to production. Their data scientists needed a platform to deploy, monitor, and retrain models without engineering bottlenecks.

The Challenge

Design an end-to-end MLOps platform that could take models from notebook to production in hours, auto-scale inference capacity, and provide real-time accuracy monitoring.

Our Solution

  • MLflow + Kubeflow Pipelines for experiment tracking, model registry, and automated training pipelines.
  • FastAPI-based inference microservices with GPU auto-scaling on AWS SageMaker for cost-efficient serving.
  • Real-time drift detection dashboard alerting the team when model accuracy degrades below a configurable threshold.
94%
Prediction accuracy
4hrs
Notebook to production
70%
Inference cost saved
12ms
p95 inference latency

Your Company Could Be Next

Tell us your challenge. We will scope the solution and give you an honest answer.

Start a Conversation