Home/How We Work

How we work

How We Build APIs

A predictable process, with the architecture agreed in writing before development starts.

The process

7 steps
1

Discovery

We start from the business objective rather than the technology, and establish what the API actually has to answer.

  • Business objective
  • Users and callers
  • Input data
  • Required output
  • Expected API volume
  • Latency requirements
  • Security requirements
  • Third-party dependencies
  • Deployment environment
2

Architecture

A written design you can review before development starts, so scope and cost are agreed on paper first.

  • Endpoints
  • Schemas
  • Databases
  • Authentication
  • Infrastructure
  • API versioning
  • Error handling
  • Scaling strategy
  • Monitoring
3

Prototype

A limited working API, so the integration can be tested against something real rather than a document.

  • Inputs
  • Outputs
  • Integration flow
  • Model behaviour
  • Performance
4

Production Development

The full build, with tests, documentation and security treated as part of the work rather than a later phase.

  • API
  • Data pipelines
  • Databases
  • Business logic
  • Models
  • Security
  • Tests
  • Documentation
5

Deployment

Deployed to an environment that suits your requirements, including your own infrastructure where preferred.

  • AWS
  • Google Cloud
  • Microsoft Azure
  • Dedicated Linux servers
  • Private cloud
  • Customer infrastructure
6

Integration

We support your engineering team through their integration, because an API is not delivered until something is calling it successfully.

7

Monitoring and Support

Once live, the service is watched continuously and supported on an agreed basis.

  • Uptime
  • Errors
  • Latency
  • Usage
  • Model health
  • Infrastructure load
  • External provider failures

Technical capabilities

Technology should support the business outcome rather than dominate it. We choose from a deliberately conventional stack, and prefer boring tools where they are sufficient.

Backend

  • Python
  • FastAPI
  • Flask
  • Django
  • Node.js where appropriate

API Standards

  • REST
  • JSON
  • GraphQL
  • WebSockets
  • Webhooks
  • OpenAPI

Data

  • PostgreSQL
  • MySQL
  • MongoDB
  • Redis
  • Object storage
  • Data warehouses

Infrastructure

  • Linux
  • Docker
  • Kubernetes where appropriate
  • AWS
  • Google Cloud
  • Azure
  • Dedicated servers

Data Engineering

  • ETL
  • ELT
  • Batch processing
  • Streaming
  • Queues
  • Scheduled pipelines
  • Data normalization

Machine Learning / AI

  • Python ML stack
  • Forecasting
  • Classification
  • Ranking
  • Anomaly detection
  • NLP
  • LLM integrations
  • Embeddings
  • RAG
  • Model-serving APIs

API quality principles

21 considerations

An API that is genuinely production-ready has to account for all of this, whether or not it appears in a proposal.

  • Reliability
  • Authentication
  • Authorization
  • Validation
  • Security
  • Data quality
  • Observability
  • Documentation
  • Versioning
  • Backwards compatibility
  • Performance
  • Latency
  • Scalability
  • Retry logic
  • Idempotency where required
  • Rate limiting
  • Error handling
  • Logging
  • Backups
  • Monitoring
  • Testing

Engagement models

MLPivot is a custom-development company, so there is no fixed public price list. Projects are scoped based on data complexity, integrations, model requirements, request volume, infrastructure and support requirements.

Fixed-scope prototype

A limited working API that proves the design before a full commitment.

Fixed-price project

An agreed scope, an agreed price, delivered against a written architecture.

Milestone-based development

Larger builds split into reviewable stages with payment per milestone.

Dedicated development team

Ongoing capacity for companies with a continuing roadmap.

Monthly API maintenance

Support, monitoring and updates for an API already in production.

Infrastructure management

Operation of the servers, deployments and monitoring around the service.

Ongoing data and model development

Continuous improvement of the data and models behind an intelligence API.

Enterprise SLA

Defined response times and availability commitments where the API is business-critical.

Want this applied to your project?

The first step is a discovery conversation about the objective, the data and the callers.