Work
Ten-plus projects delivered. One of them is named — Nucleus, because it’s ours to talk about. The rest are here as what we built, not who we built it for.
Our clients are large, and their names aren’t ours to publish. The work is. Each project below is the system, the stack it runs on, and what it does now that it didn’t before.
If you want more than a page can carry — the constraints we designed around, how the team was staffed, what we’d do differently — ask on the call. We’ll go as deep as the agreement allows, which is usually deeper than you’d expect.
How the work runs
A short list of case studies, and most of what we’ve built sits under agreements that don’t allow a page like this one. So here is the part that doesn’t change from one project to the next.
-
The engineer who scopes it builds it. You meet the people who’d do the work on the first call, not an account manager who hands you on afterwards. You will know everyone on your project by name, and nobody is learning the basics on it.
-
A demo every week, in writing. What moved, what didn’t, what we decided and why — every week, whether or not the week went well. You should never have to ask how it’s going.
-
The estimate is a range, with its assumptions attached. We quote a band and say what it rests on. When something invalidates an assumption you hear it the week it happens, not in the final invoice.
-
It's yours from the first week. Code, infrastructure and documentation live in your accounts from the start, not ours. There is no handover event, because there is nothing to hand over.
-
We're still here after delivery. If something we shipped breaks, it’s ours to fix. Past that, a retainer if you want us on call and nothing if you don’t — we don’t bill for a relationship you aren’t using.
-
Ask about the rest. The projects that aren’t on this page, the engineers who’d be on yours, where we’ve done this in your sector — ask on the call. At proposal stage we’ll put you on the phone with someone we’ve worked for.
What we know going in
Every project in a discipline rhymes. These are the questions clients arrive with, and the work that follows them — so you can judge the thinking before you commit to the invoice.
Data engineering and analytics
-
Change data capture and ingestion
Your Fivetran bill has doubled on monthly active rows? Need CDC off SQL Server or Postgres without a Kafka cluster to babysit, and can’t tell whether Debezium, dlt or a managed connector is cheaper over three years? We’ll cost it out and build it.
-
Medallion lakehouse build
Building a lakehouse and stuck on what belongs in bronze, silver and gold? Still arguing over S3 or ADLS Gen2, Iceberg or Delta, Spark or just a warehouse? Not sure your volumes justify Databricks at all? We’ll design it around your data.
-
Streaming with Kafka and Flink
Someone has asked for real-time and nobody has defined it? Wondering whether Kafka and Flink are warranted or whether micro-batches every five minutes would do, and who handles late events, replays and exactly-once? Let’s size it honestly first.
-
dbt and SQLMesh modelling
Business logic buried in stored procedures on a SQL Server that is now out of support? Hundreds of dbt models, no tests, and incremental runs you don’t trust? Weighing up a move to SQLMesh? We’ll untangle the layer and document it.
-
Orchestration and on-call reliability
Pipelines failing nightly with nobody on call? The Airflow 3 upgrade keeps slipping, or you’re asking whether Dagster is a better home. Nobody knows how the triggers and backfills work? We’ll hold the pager while you learn.
-
Data quality and contracts
Finance spotted the bad numbers before you did? No freshness checks, no owner for a broken table, and an upstream team that changes schemas without telling you? Unsure whether Soda, dbt tests or contracts fit? We’ll put checks where they pay.
-
Governance and PII access control
Auditors asking who can read customer PII and you can’t answer in writing? Everyone sits on one admin role, row filters and column masks are undocumented, Unity Catalog or Purview half configured? We’ll tag what is sensitive and prove access.
-
Warehouse cost reduction
Snowflake credits climbing every month with no new users and no new data? Warehouses that never suspend, every team sharing one, and nobody able to say which dashboard burns the money? Databricks DBUs the same story? We’ll find the spend and cut it.
-
Semantic layer and Power BI
Three Power BI reports, three versions of revenue? DAX nobody dares touch, imports refreshing for two hours, and an open question about whether metrics belong in dbt or Cube instead? We’ll define each measure once and move reporting onto it.
Software development
-
MVP and prototype builds
Got a launch date and investors but no engineering team? Not sure what belongs in version one, whether to start on web or mobile, or if Next.js and Postgres will carry you past launch? We scope it, then build it.
-
React and Next.js web apps
Stuck halfway through an App Router migration? Hydration errors that only show up in production, server components nobody on the team understands, and an INP score that’s hurting you? We’ll finish the move.
-
Cross-platform mobile apps
Need one app on iOS and Android? Can’t choose between React Native, Expo and Flutter, or your build was rejected for App Completeness and the rating is sinking under crash reports? We ship and resubmit.
-
API and backend rebuilds
Is your backend one monolith and the person who wrote it has left? Need to version a REST API without breaking mobile clients, add webhooks and rate limits, or move services out one at a time? We pull it apart safely.
-
Database design and performance
Queries that were fine at ten thousand rows now timing out? Nobody reading pg_stat_statements, and no idea whether you need indexes, PgBouncer, read replicas or a different schema? We measure before changing anything.
-
Cloud infrastructure and CI/CD
Deploys still done by hand on a Friday? No Terraform, no staging that matches production, a CI pipeline slower than code review, and an AWS bill nobody can explain line by line? We put the pipeline in.
-
Payments and subscription billing
Moving to usage-based pricing and unsure how Stripe’s meters fit what you already built? Need proration, dunning on failed cards, tax handling and webhooks you can trust to be idempotent? We own the money path.
-
Auth, SSO and compliance
Lost a deal because you have no SAML or SCIM? Need to migrate off Auth0, scope MFA and audit logging ahead of a SOC 2 Type II window, or clear a pentest finding list before renewal? We do the remediation.
-
Web3 and on-chain features
Adding wallet sign-in? Not sure how to verify an EIP-4361 signature server-side, where the session token then comes from, or whether to read chain state over RPC or index it with a subgraph? We’ll wire both up.
AI and automation
-
RAG chatbots on your data
Want staff to ask questions of your own documents instead of hunting through SharePoint? Worried the bot will answer from the wrong file, or show someone a contract they shouldn’t see? We build the retrieval and the permission checks.
-
Retrieval quality tuning
Your assistant answers confidently and gets it wrong? Chunks split mid-table, no idea whether to add BM25 alongside embeddings, or whether a reranker would fix it? We measure retrieval first, then tune what’s actually broken.
-
Agentic workflow automation
Support tickets triaged by hand, invoices matched to POs by hand? Tried an agent that worked in the demo and fell over on the edge cases, with no path for exceptions? We scope the steps worth automating and leave humans the rest.
-
MCP servers for your systems
Need your agents reading from Salesforce, ServiceNow or your own Postgres? Unsure which tools to expose, how to scope auth per user, or how to stop a poisoned tool description? We write the MCP server and run it for you.
-
LLM evaluation harnesses
No way to tell whether last week’s prompt change made things better or worse? Shipping on vibes, nothing in CI, no record of what the model said in production? We build the eval set, wire up Langfuse or promptfoo, and gate releases.
-
Self-hosted open-weight models
Legal says the data can’t leave the country, or can’t reach a US API at all? Weighing Llama, Qwen and Mistral licences, sizing GPUs, unsure whether vLLM on your own hardware will hold the load? We size it and run it.
-
Document and invoice extraction
Thousands of invoices, statements or forms keyed in by hand every month? Textract reads the text but not the fields, and an LLM invents a total when it can’t find one? We add validation rules and confidence thresholds for review.
-
Model choice and cost control
API bill tripled and nobody knows which feature caused it? Not sure if prompt caching, a cheaper model for the easy calls or the batch API is the fix, or whether you need fine-tuning at all? We measure cost per task first.
Tell us what’s broken.
Twenty minutes, no deck. We’ll tell you straight whether this is work we should be doing, and if it isn’t, we usually know who should.
We come back within one business day.