Anthony Jeyifous

Data and Analytics Engineer, Berlin

I build data platforms people can trust, from industrial sensors to dashboards.

Five years as the only data and automation engineer on a 3D concrete printing research rig at BAM, Germany's federal materials research institute. Today I own a production sports data platform and its observability layer. M.Sc. Chemical and Energy Engineering, B.Eng. Mechanical Engineering, Ironhack Data Engineering.

CV on request GitHub

Arbibet platform in numbers

as of 2026-10-21 23:17 UTC · status page

204 k

records ingested, last 24 h

99.3%

uptime, last 7 days

under 1 min

age of newest data at that time

What I work on

Data engineering

Ingestion from 20+ external sources and APIs with checkpointed runs, dead-letter handling and content-hash deduplication, covered by over 200 automated tests.
Lossless storage for high-frequency machine data: raw logs into Parquet with session manifests and schema-drift capture, up to 6.4× smaller, queried with DuckDB.
Tools chosen by measurement: a Snowflake, Airflow and Kafka pipeline moved to DuckDB and an event-driven runner, with identical results, latency from up to an hour to seconds, running cost from $212 of trial credit to $0.

Analytics and data modelling

Star-schema dimensions and partitioned fact tables fed by incremental loads; dbt models with tests.
One metric contract behind every report and status page, so each number has one definition.
Process KPIs derived from machine telemetry, and parameter sweeps that turned test data into decisions.

Observability and reliability

Read-only monitoring across 4 production databases that separates data freshness from pipeline lag.
Production incidents driven to durable fixes: "succeeded" runs that refreshed nothing (a stale image), an archive switch taken through out-of-memory incidents and a tested rollback.

Process engineering

Parameter sweeps that defined a stable printing window and cut failed prints, where one failed print cost a day of material and rig time.
A mass balance for a continuous mixer with no solids measurement, from timed nozzle weighings and a stirred-tank model, giving an online water-to-solids ratio.
Water-valve-to-torque dynamics identified from routine production data, and PI versus model-predictive water control compared in a simulator that replays real prints (simulation).

Mechanical engineering and instrumentation

A pumping line instrumented with pressure and temperature transmitters, a vortex flow meter and a proportional water valve, with 4 to 20 mA loop fault detection.
The rig's support grillage designed in Siemens Solid Edge and machined; test rigs for carbon-fibre yarn built at SGL Carbon (CATIA V5).

Projects and operations

A research rig taken from concept and component specification through build, commissioning and handover; supplier offers compared and deliveries accepted.
About EUR 30,000 saved by replacing a quoted commercial control package with an in-house Python, OPC UA and Node-RED stack.
Operators trained, runbooks and checklists written, statistical process control with material QC tests.

Projects

All projects

Arbibet: sports data platform and its operations

A production sports data and market-signal platform I architect and run: ingestion from 20+ sources, a star-schema warehouse on PostgreSQL, a Parquet cold archive queried with DuckDB, and its own observability layer. I write the architecture, specs and acceptance criteria; implementation runs through a spec-driven, AI-assisted workflow where an independent verifier closes every task.

BAM: data and control for a 3D concrete printing rig

Five years as the only software and data engineer on a research rig: real-time OPC UA ingestion across the PLC network, an in-house control stack, closed-loop bead-width control, statistical process control, and 5+ publications (2 as first author). Since May 2026, as a guest researcher, I rebuilt its acquisition and storage layer: sensor ingestion up 5×, lossless Parquet storage, and a private API serving layer.

AI Fragrance Advisor (freelance, for a fragrance retailer)

A bilingual advisor that turns a customer's answers into three picks from the retailer's own range, with privacy by default and evaluation gates.

Demo on request.