Development

The part where strategy becomes software people can actually use.

We still love a good interface. But a lot of the work that matters most happens behind the scenes: APIs, orchestration, data pipelines, evaluation logic, analytics, integrations, and the systems that keep everything steady once real users arrive.

We’re not religious about stacks. We use the right tools for the job. That often means modern frontend work on one side, and Python on the other when a product needs heavier backend logic, data work, scientific computing, model integrations, or architecture built to grow.

Python, APIs, and backend systems

We build production-grade Python services end to end: REST and GraphQL APIs, background jobs, event-driven workflows, integrations, and deployments across cloud or on-prem environments.

The goal is not just to get the thing live.
The goal is to make it maintainable.

That means automated testing, sensible typing, CI/CD, structured logging, metrics, tracing, and code another team can understand without needing the original developers to translate it six months later.

Still friends with the frontend

We also know our way around React, Next.js, TypeScript, and interface-heavy products. That matters when the job is not just building a backend, but building the seam between the backend and the humans using it.

Right tools. Right boundaries. Less pain later.

LLMs, evaluation, and AI systems

We build LLM-enabled applications for the real world, not just demos that survive one stakeholder meeting.

That includes retrieval pipelines, RAG, tool and function calling, structured outputs, human review loops, safety guardrails, privacy considerations, and production monitoring. We care about measurable quality: benchmarks, evaluation flows, latency, cost, failure modes, and how systems behave when models get things wrong.

When a system needs to support multiple providers, model types, or output formats, we design for extension from the start. Adding a new model should feel like a planned integration, not emergency surgery.

Data, analytics, and visualization

Some products are mostly screens. Others are data products wearing a nice coat.

We work across ingestion, cleaning, transformation, analysis, and visualization. We use the scientific Python ecosystem when it makes sense, build reproducible workflows, and treat reporting and dashboards as part of the product rather than something bolted on at the end.

Clear data models. Repeatable pipelines. Fewer mystery numbers.

From monolith to maintainable platform

A lot of products start as one big app because that’s the fastest way to learn. Fair enough.

The trouble starts when that first version has to support more users, more workflows, more integrations, and more teams. Then the work becomes separating concerns without losing momentum.

We help untangle monoliths, define cleaner boundaries between frontend and backend, preserve what already works, and rebuild the parts that are slowing the whole system down.

Long-term maintenance and open source

Shipping version one is one job. Keeping software healthy is another.

We care about documentation, release discipline, contribution workflows, and the boring but essential engineering work that makes software usable by teams beyond the original builders. Good open source needs clear boundaries, clear docs, and maintainers who take the community side of the work seriously.

When a product needs both an open core and a managed version, we prefer clean separation over clever mess.

Built for real-world constraints

We’ve been building digital platforms since the late 1990s, which is long enough to learn that not everybody has the latest device, perfect connectivity, or endless engineering support.

We optimize for older hardware, uneven bandwidth, operational constraints, and teams that need software to be understandable, stable, and cost-conscious. That matters in every product, and even more in public-interest, research, and civic-tech work.

Public-interest digital work

We’ve contributed to public-interest digital work across Hungary, Ukraine, Malawi, Zimbabwe, Namibia, and Morocco — from transparency and public budget monitoring to parliamentary participation and open access to extractives data.

That work has included collaborations with organizations such as The Engine Room, SimSim, and IPPR, across projects involving public budget monitoring, replication work in Malawi and Zimbabwe, Transparent Oil Namibia, and Nouabook in Morocco.

These are different contexts, but the underlying challenge is often the same: make important systems easier to understand, easier to question, and easier for people to use. That takes practical technology, careful information design, and software that holds up when the environment is messy, bandwidth is limited, or the audience extends well beyond a technical team.