00Mission gateway
Michel
Fernandy
Eloka
Data Engineer
Move traceable data from source to decision.
- Python
- Kafka
- Spark
- Airflow
- MinIO
- PostgreSQL
5source families3 zonesRAW · CURATED · DATAMART865AI answers evaluated
Available immediately · Paris / Île-de-France
Scroll to enter the system
01Experience & evidence
Data should drive action.
In banking, I built monitoring dashboards. On SIRIUS, I worked on a complete chain from public sources to BI.
2 contextsbanking + engineering
Objects to explore · 3Click an object in the scene or in the list.
- Banking Power BI Professional contribution · Crédit Coopératif
- Built Power BI dashboards and automated controls/workflows with Power Automate. More visible indicators for teams.
- HRIS data Professional contribution · Crédit Coopératif
- Preserved business rules and helped secure data reliability. A web migration without losing the data’s business meaning.
- Public sources Engineering project · team of 3
- Python collectors for Bluesky, Nostr, Hacker News, RSS and Stack Overflow. Five source families normalised for ingestion.
02Global interaction
Follow my data.
Launch an event. At each zone, the scene reveals what I implemented and why the stage exists.
6 stagesfrom source to BI
Objects to explore · 6Click an object in the scene or in the list.
- Collect Engineering project · team of 3
- Developed Python collectors and normalised events. A coherent input format before ingestion.
- Kafka + Redis Engineering project · team of 3
- Kafka publishing and Redis-based deduplication. An asynchronous flow more robust to duplicates.
- MinIO RAW Engineering project · team of 3
- Stored raw data in MinIO as JSONL. A retained source before transformation.
- Spark + Airflow Engineering project · team of 3
- Airflow for orchestration and Spark for transformations into Parquet. Ordered processing and a usable curated zone.
- Serve Engineering project · team of 3
- ClickHouse staging followed by a PostgreSQL datamart. A structured layer adapted to reporting.
- Visualise Engineering project · team of 3
- Reporting with Streamlit, Superset and Power BI. Exploration and dashboards on prepared data.
03SIRIUS Data architecture
Three zones, one lineage.
The pipeline separates raw retention, controlled transformation and analytical serving.
RAW → CURATED → DATAMARTdocumented pipeline
Objects to explore · 4Click an object in the scene or in the list.
- RAW Engineering project · team of 3
- MinIO and raw JSONL files. Replayability and source traceability.
- CURATED Engineering project · team of 3
- Spark jobs orchestrated by Airflow, Parquet output. A layer ready for analytical uses.
- DATAMART Engineering project · team of 3
- Modelled and served data in PostgreSQL. A queryable base for reporting tools.
- Decision Engineering project · team of 3
- Streamlit, Superset and Power BI dashboards. The chain ends in interpretation, not just storage.
04Incident S4 & experiment
When Kafka goes silent, where do you look?
Scenario S4 follows producer–topic–consumer. The thesis extends that rigour through a campaign where failures are not hidden.
900 / 865 / 35planned · usable · failures
Objects to explore · 4Click an object in the scene or in the list.
- Kafka chain Individual work
- Checked producer, topics, consumer lag and logs before diagnosis. Locate the break point instead of concluding too early.
- Quality Individual work
- Retained 35 failures or empty outputs with a zero score. A reliability measure without cherry-picking.
- Evaluate Individual work
- Scored quality, evidence, stability, latency, tokens and hallucinations. A decision grounded in multiple dimensions.
- Decide Individual work
- Compared LLM, RAG and agent approaches across two models. RAG by default, agent on selective escalation.
05Beyond the pipeline
Measure without losing meaning.
Mathematics, tutoring and chess reinforce my analytical rigour — and my ability to explain.
Objects to explore · 3Click an object in the scene or in the list.
- Teach Individual work
- Adapt explanations to each student’s level, blockers and pace. Eight students supported and sustained teaching practice.
- Anticipate Individual work
- Organise, facilitate and hand over responsibilities after graduation. Chess strengthens how I read sequences, consequences and trade-offs.
- Connect Professional contribution · Crédit Coopératif
- Document decisions, make progress understandable and share risks. A dual BUILD/RUN culture and a common language across stakeholders.
06Next mission
Let’s move data we can explain.
This route prioritises ingestion, orchestration, quality, datamart and BI; its résumé is built for a Data role.
Explore another trajectory