Engineering capacity, with the judgement attached

AI and LLM Engineering

Retrieval, agents and copilots that hold up under real usage.

• LangChain
• LangGraph
• OpenAI
• Anthropic
• pgvector
• Pinecone

We build LLM features the way we build the rest of the product: versioned, tested and measured. That means an evaluation suite before launch, retrieval you can trace back to a source document, and cost and latency budgets agreed up front rather than discovered on the first invoice. Most of our work here runs on LangChain and LangGraph, with the orchestration layer kept thin enough to swap a model provider in an afternoon.

• RAG pipelines with citation-level provenance
• Multi-step agents and tool use on LangGraph
• Evaluation harnesses and regression suites
• Model routing, caching and cost controls
• Guardrails, PII redaction and audit logging
• Fine-tuning and small-model distillation

Product Engineering

Full-stack teams that take a product from scope to production.

• TypeScript
• React
• Next.js
• Node.js
• Python
• PostgreSQL

A pod of three to six engineers with a tech lead who stays on the project start to finish. We work in your repositories, in two-week increments, with a demo at the end of each one. You are never more than ten working days from something you can click on, and never dependent on a person you have not met.

• Greenfield web and mobile products
• API design and third-party integrations
• Design systems and component libraries
• Payments, billing and subscription flows
• Role-based access and multi-tenancy
• Automated testing and CI/CD pipelines

Data and Platform

The layer everything else depends on, built to stay boring.

• Airflow
• dbt
• Snowflake
• Kafka
• Terraform
• Kubernetes

Ingestion, warehousing and the contracts that keep schema drift from quietly corrupting a downstream model. We also handle the infrastructure underneath: Terraform-managed environments, sensible observability, and deploys that a single engineer can run on a Friday without a war room.

• Batch and streaming ingestion pipelines
• Warehouse and lakehouse modeling
• Data contracts and quality monitoring
• Infrastructure as code on AWS, GCP and Azure
• Observability, alerting and on-call runbooks
• Cost review and rightsizing

Legacy Modernization

Getting an old system moving again without stopping the business.

• .NET
• Java
• PHP
• Python
• Docker
• PostgreSQL

Most of the systems we are called into cannot be switched off for a rewrite. We work incrementally. Carve out a bounded piece, put it behind a stable interface, replace it, then repeat, so value lands every increment and there is never a single terrifying cutover weekend. We have done this with monoliths well past a decade old, and with vendor systems nobody on staff had shipped to in years.

• Strangler-fig migrations off monoliths
• Framework and runtime upgrades
• Database migration and dual-write cutovers
• Test harnesses for untested code
• Documentation of undocumented systems
• Security and dependency remediation

Team Extension

Vetted senior engineers embedded in your team, not a black box.

• Time zone overlap
• Direct Slack access
• Your tooling
• Your process

When you need capacity rather than a project, we place engineers from our network directly into your team, your standups and your code review. Every candidate has been through our own technical screen and a paid trial task before you see them. If a placement is not working in the first two weeks, we replace them and you are not billed for the overlap.

• Senior full-stack and backend engineers
• ML and data engineers
• DevOps and platform engineers
• Mobile engineers (iOS, Android, React Native)
• QA automation engineers
• Fractional tech leads and architects

Discovery and Architecture

A two-week engagement that ends in a plan you could hand to anyone.

• 2 weeks
• Fixed price
• Written deliverable
• No obligation to continue

Before a large build we run a short paid discovery: we read the code, interview the people who maintain it, and come back with a written architecture, a risk register and a costed roadmap. Roughly a third of these end with us recommending a smaller project than the client came in asking for. That is a good outcome and we would rather find it in week two than month six.

• Codebase and architecture review
• Technical due diligence for investors
• Build-versus-buy analysis
• Costed and sequenced roadmap
• Risk register with mitigations
• Fixed-price proposal for the build