20.06.2026 aktualisiert


100 % verfügbar
Senior Backend & Data Engineer | Python, Django, PostgreSQL, AWS | RAG/LLM Integration
Geneva, Schweiz
Weltweit
MSc Civil & Hydraulic Eng (Numerical Modeling), Grenoble INP-ENSE3. MEng ENSH (valedictorian)Über mich
I turn research code and proofs of concept into tested, deployable Python backends. Six years across a Swiss catastrophe-risk platform, hydropower forecasting, and applied research, often owning delivery end to end as engineer and product owner.
Skills
DagsterGitHub ActionsAPIsApache AirflowLinuxDevOpsDjangoPythonPostgreSQLPHP DevelopmentPostgisRabbitMQAWS CLIDatenbank-EnginesFlask
Senior backend and data engineer with 6+ years building and shipping production Python systems for enterprise and research clients across France and Switzerland.
Backend and APIs: Python (advanced, OOP), FastAPI, Pydantic, Django, Django REST Framework, Flask. REST API design, OpenAPI, asynchronous task queues with Celery and RabbitMQ, microservices, ORM and query optimization (N+1 elimination, PostgreSQL and PostGIS tuning on 10M+ records).
Data engineering: end-to-end ETL pipelines in Python, multi-source API ingestion, Dask and xarray for distributed processing, orchestration with Dagster and Airflow, Parquet and cloud-optimized formats, PostgreSQL and PostGIS (GiST indexing, materialized views, partitioning).
AI and LLM: integrating LLM APIs (OpenAI, Claude, Mistral, AWS Bedrock) into typed Python backends, and building retrieval-augmented (RAG) services on FastAPI and pgvector with hybrid retrieval and reranking. Strong fit from a data-engineering background: retrieval quality, chunking, evaluation and data pipelines are the hard parts of RAG, and these are problems I have solved for years.
Cloud and DevOps: AWS (S3, EC2, Lambda), Docker, Docker Compose, Terraform, CI/CD with GitHub Actions and GitLab CI, Linux, SLURM and HPC.
Background spans a Swiss catastrophe-risk platform serving insurers, hydropower forecasting in the energy sector, and applied research at a national institute. Comfortable turning research code and proofs of concept into tested, documented, deployable services.
Erfahrung
Berufserfahrung5 - 10 Jahre
Freelancer seit2025
Sprachen
ArabischMutterspracheEnglischMutterspracheFranzösischMuttersprache
Projekthistorie
Backend and data engineering for early-stage software and environmental-data products.
- Designed and shipped a FastAPI service over PostgreSQL/PostGIS, with a Dagster-orchestrated data pipeline decoupled from the API to keep request latency stable under heavy compute.
- Provisioned reproducible AWS infrastructure (S3, compute, networking) as code with Terraform, replacing manual setup with versioned environments.
- Delivered a containerized platform (Django, PostGIS, cloud-optimized formats) to the client's own infrastructure with Docker Compose.
- Built a retrieval-augmented (RAG) service on FastAPI and pgvector for natural-language search over technical document sets.
Industrialized a flood-risk research prototype into production geospatial tooling, deployed to 15+ field users.
- Refactored research notebooks into a modular, tested Python package (pytest, property-based tests) with clear API contracts, now used by the research team.
- Built Dask and xarray pipelines on SLURM, with API-driven ETL from Copernicus, IGN and Theia into cloud-optimized formats, backed by a partitioned, GiST-indexed PostGIS schema.
- Contributed to a multi-user platform for satellite-data upload, cataloging and processing, integrating OpenEO datacube workflows.
- Diagnosed multi-stage pipeline failures with structured logging, isolating raster corruption, CRS mismatches and silent feature drops.
Production backend and data engineering on a global catastrophe-risk platform serving institutional and insurance clients.
- Converted climate scientists' research code into a production analytics backend, taking hazard and impact models from scripts to tested, deployable services with clean APIs.
- Built cloud-native ETL converting multi-terabyte vector and raster archives into Parquet and cloud-optimized formats on AWS S3 with STAC metadata, forming a queryable data lake.
- Served hazard and exposure layers through FastAPI over PostGIS, cutting spatial-join latency from 5 minutes to 2 seconds via GiST indexing and materialized views.
- Processed large climate archives (CMIP6, CORDEX, ERA5) on SLURM across 500+ parallel jobs, with uncertainty quantification (Monte Carlo, Sobol).