UK-based technology consultancy
Engineering systems where the cost of failure is measured in millions.
Redcliffe Digital is a small team of senior engineers, architects and SREs. We learned our craft building trading systems where a five-second outage is a board-level incident. We now apply that discipline to public sector technology.
Data platform & warehouse engineering
Data platforms engineered for when the numbers have to be right.
Redcliffe Digital is a small team of senior engineers building the warehouses, pipelines and data systems organisations depend on, modelled for correctness and built to scale. We learned the craft moving sub-millisecond market data, where a wrong number costs millions, and we bring that discipline to everything we build.
Accreditations and partnerships
AWS Certified Developer – Associate
Microsoft Certified: Azure Fundamentals
Cyber Essentials
ISO 27001
Disability Confident Committed
What we do
Platform engineering
Cloud-native systems on AWS and Azure, built to the GDS Service Standard and the Technology Code of Practice. We do the unglamorous parts well: observability, infrastructure as code, incident response runbooks.
Trading-grade resilience
We design for the long tail. Disaster recovery, chaos testing, capacity planning, and the kind of pre-mortems that surface the failure mode no-one wanted to talk about.
Delivery in the open
Small, embedded teams that work in the open alongside civil servants. We leave you with running software, documentation, and a team that no longer needs us.
Data warehouse engineering
Warehouses and lakehouses on Snowflake, BigQuery, Redshift and Databricks: modelled for correctness, partitioned for scale, and governed for regulated and sensitive data. Every published figure traces back to its source row.
Platform engineering
Cloud-native systems on AWS and Azure, built to the GDS Service Standard and the Technology Code of Practice. We do the unglamorous parts well: observability, infrastructure as code, incident response runbooks.
Trading-grade resilience
We design for the long tail. Disaster recovery, chaos testing, capacity planning, and the kind of pre-mortems that surface the failure mode no-one wanted to talk about.
Our core practice
How we build a data warehouse
A warehouse is judged on three things, and we engineer each as a first-class requirement: that it stays reliable when a source misbehaves, that it stays scalable as volume grows, and that it stays maintainable for the team that inherits it. Most we are asked to rescue failed on the last one long before the others.
Ingestion that does not lie
Log-based change-data-capture for freshness, idempotent batch loads where nightly is right. Every load is replayable, so a bad transformation is a redeploy, not a data-loss incident.
Modelling for one answer
Conformed dimensions and facts in the Kimball tradition, layered raw → core → marts. Two analysts asking the same question get the same number, every time.
Lineage end to end
Version-controlled, tested dbt transformations and automatic lineage. Any published figure traces back through the model to the exact source row that produced it.
Scale and cost held flat
Partitioning and clustering that keep tens-of-terabytes queries fast, with cost attribution so a per-query bill never becomes a nasty surprise.
Governance by design
Classification at ingestion, least-privilege auditable access, and retention and erasure enforced by the pipeline, so a deletion request provably reaches every derived table. Built for regulated and sensitive data from the first sprint.
Schema evolution, not breakage
Explicit data contracts, backward- and forward-compatible encodings, and migrations tested against real historic data. Sources change; the warehouse bends rather than breaks.
Why a small firm beats a big one
The Big Four can field a hundred consultants by Monday. We can’t. What we can do is put two principal engineers in a room with your team on day one, the same people who will still be there in month nine. There is no offshore handover, no rotating bench, no upward delegation. The person you meet at the kick-off is the person writing the code.
- Senior-only delivery teams. No pyramid.
- Decisions made in the room, not escalated.
- Day rates 30–40% below tier-one consultancies.
- We turn down work we cannot do well.
Case studies
Selected work
Cyber security and AI
A self-hosted RAG pipeline that took threat attribution from hours to seconds
A threat intelligence firm could not put sensitive feeds through commercial LLM APIs. We built a self-hosted retrieval-augmented pipeline that answers attribution queries in seconds, with provenance the analysts trust.
Read case studyFinancial services / Quantitative trading
Replacing a C# monolith with a low-millisecond market-data platform
A quant firm’s market-making stack was a monolith that throttled every new idea. We migrated it to Python microservices and built a Rust market-data layer running at low-millisecond latency across multiple exchanges.
Read case studyAcademic research / Data & analytics
An R/Shiny analytics tool that put behavioural research data in non-technical hands
The School of Psychology at Keele University had large volumes of behavioural and observational data and no way for non-technical staff to interrogate it. We designed, built and deployed an interactive R/Shiny tool, co-designed with the researchers and now used by 15+ staff.
Read case studyCyber security / Data & analytics
Consolidating fragmented detection telemetry into one governed threat-intelligence warehouse
A managed threat-intelligence provider ran its detections and customer reporting off a dozen disconnected telemetry stores. We built a single governed warehouse (modelled, lineage-tracked and tenant-isolated) that answers cross-source threat questions in seconds.
Read case studyFinancial services / Quantitative trading
Replacing a C# monolith with a low-millisecond market-data platform
A quant firm’s market-making stack was a monolith that throttled every new idea. We migrated it to Python microservices and built a Rust market-data layer running at low-millisecond latency across multiple exchanges.
Read case studyCyber security and AI
A self-hosted RAG pipeline that took threat attribution from hours to seconds
A threat intelligence firm was drowning in terabytes of heterogeneous feeds it could not put through commercial LLM APIs. We built the governed data platform underneath (high-volume ingestion, normalised storage, full provenance) and a self-hosted RAG layer that answers attribution queries in seconds.
Read case studyAcademic research / Data & analytics
An R/Shiny analytics tool that put behavioural research data in non-technical hands
The School of Psychology at Keele University had large volumes of behavioural and observational data and no way for non-technical staff to interrogate it. We designed, built and deployed an interactive R/Shiny tool, co-designed with the researchers and now used by 15+ staff.
Read case studyWorking on a programme where failure is not an option?
Tell us what you’re trying to deliver. We’ll tell you honestly whether we can help.