Equipping teams without taking their decisions for them.
Six services with one requirement: everything an engagement produces must be readable, reproducible and approvable by an in-house engineer. Without a handover procedure the client inherits files it cannot use. A generated deliverable nobody can explain causes the same problem.

The technologies used in production and the role of each. The choice depends on the client’s constraints. A stack the in-house team cannot maintain ends up being replaced, at its own cost.
Storage, query, transformation and reporting. The choice depends on the client’s constraints: an existing cloud contract, sovereignty requirements, in-house skills.
A SQL query engine that reads files stored on S3 directly, with no database to load and no server to maintain. Billing depends on the volume of data scanned, which makes partitioning a design decision to take from the start.
Hadoop, Spark and Trino clusters created on demand, for workloads too large for a single machine. Existing Spark code runs there without a rewrite, which is often the reason for choosing it.
Azure’s serverless service: code triggered by an uploaded file, a schedule or an HTTP call, with no server to size. Suited to short processing steps; for long-running jobs, per-second billing makes it expensive.
Microsoft’s relational engine, widespread in enterprise systems. In a migration it is almost always the source to read and the constraint to respect, rarely the target.
The default target when nothing requires a proprietary engine: open licence, portable across clouds. Its extensions, PostGIS for geospatial data and pgvector for vector search, avoid adding a second system for a single need.
The GPT models, called through OpenAI’s API or through Azure OpenAI when the client already has an Azure contract. Used for extraction, classification and writing assistance. Engineering decisions stay with an engineer.
Anthropic’s Claude models, called through Anthropic’s API or through AWS Bedrock, Google Vertex AI and Microsoft Foundry. The client’s cloud provider therefore does not dictate the choice of model. Used for document analysis, writing assistance and agents that call tools.
Two products under the same name: GitHub Copilot for code assistance in the editor, and Microsoft Copilot with Copilot Studio for business agents on Azure. The agents built at Egis Group are based on the latter.
A data warehouse that separates storage from compute: each is sized separately, and compute is billed per use. Available on all three major clouds, which avoids tying the warehouse to the infrastructure provider.
Several model families available through a single API, without data leaving the AWS account. The choice of model is recorded as an architecture decision, open to revision.
Ingestion, warehouse, notebooks and reporting in one product, on shared storage (OneLake). The integration saves time; in return, the platform ties the client to the Microsoft ecosystem. That trade-off is settled at scoping.
Semantic models, reports and dashboards, used directly by business teams. Ambiguous metric definitions become visible there, which makes it a good test of the quality of the data model upstream.
Three languages, chosen because an in-house team can read them with no special training.
The language of data processing and machine learning, from the ingestion script to the model service. pandas, PySpark and scikit-learn cover every step without changing tools.
The only language read by both business and technical teams. A transformation written in SQL stays readable years later, which is rarely the case for a pipeline written in code.
Describes cloud infrastructure in versioned files, applied reproducibly. Without it, a platform’s configuration exists only in a web console, where it cannot be reviewed, handed over or audited.
Two providers used in production. Google Cloud is covered by certifications but not listed here, for lack of a recent engagement to cite.
The widest catalogue of data services: S3, Athena, EMR, Bedrock. It is the default choice when the company does not already have a Microsoft contract.
The natural provider for companies already on Microsoft 365 and Entra ID, where identity, billing and compliance are already in place. Fabric, Functions App and Azure OpenAI are its most common data services.
Every engagement follows the same four steps. Decisions are taken at scoping, and that is the step that is tracked. The scope is written down so that nobody waits for an access, an environment or a decision.
Requests declined, whatever the length of the engagement.
- An engineering decision automated with no accountability framework.
- A deliverable no in-house engineer can read.
- A figure announced without a measurable source.
- Scaling up before a pilot has run in production for three months.