Hamza A

• Designed the architecture of Limagrain's internal agentic platform on Google ADK / Vertex AI, supporting daily operations of the group's Data Platform (ODP): a multi-agent system (specialized agents for proposing, judging, reducing, verifying, and reporting incidents) that detects issues across 22+ monitored services (Dataform, Fivetran, Eventarc, BigQuery), classifies root causes (infra / code-bug / permission), correlates failures with recent commits, and generates actionable remediation steps. • Includes Compass, an interactive ADK agent sharing the same tool surface and grounded in • ODP documentation, runbooks, and naming / governance best practices (RAG), wrapped in a • Next.js + CopilotKit chat UI — lets the data platform team diagnose incidents, get guidance on conventions, and unblock themselves on best practices without escalating. • Owned the LMC Finance data product end-to-end: designed and shipped Dataform pipelines, Cloud Functions, Cloud Run jobs, and the PowerBI semantic model feeding finance reporting — full ownership from ingestion to consumption. • Collaborated with the data platform team to roll out AI-assisted ops practices, reducing manual incident triage time. Stack: GCP, BigQuery, Dataform, Fivetran, Terraform, Google ADK, Vertex AI, Gemini, RAG, CopilotKit, Pub/Sub, Next.js, Python, Cloud Run, Cloud Functions, PowerBI, Dataplex

• Designed and shipped end-to-end data & AI platform on GCP (BigQuery, Dataform, BQML, • Fivetran, Eventarc, Cloud Workflows): 30+ refined/golden tables, Eventarc-triggered Cloud • Workflows orchestrating Dataform runs and 12 downstream Google Sheets refreshes on every Fivetran landing, CI/CD with gated compile + unit tests, Dataplex data quality and catalog —dashboard freshness reduced from 24h to 3h. • Productionized a predictive churn model (BQML XGBoost, AUC 0.90): 50+ behavioral,engagement and transactional features engineered in-database, with segment predictions validated against ground truth. • Extended the ML portfolio on a shared feature store: unsupervised segmentation (BQML Kmeans) for marketing cohorting, and NLP-based frustration detection using LLMs on call transcripts and support tickets to surface at-risk members early. • Built Blast Copilot, a conversational analytics layer on Gemini in BigQuery with a Dataplex business glossary and verified-SQL guardrails for Sales and Marketing. Delivered the reporting and dashboard layer: multiple Metabase dashboards (engagement, app study, churn, retention, no-invest year 1) plus auto-refreshed Google Sheets connected to BigQuery for operational CRM teams — removed manual reporting work and ensured consistent metric definitions across teams. • Implemented end-to-end data governance: column- and row-level security via Dataplex policy tags, PII masking for analytics dashboards, GDPR-compliant access patterns.

• Working on designing and building data products around Kering’s customer domain, using a data mesh centric approach with DBT, BigQuery, and SQL. • Also maintaining legacy data lake on AWS, and handling the maintenance and migration of core Spark scala pipelines • Built the conformed Customer Dimension on BigQuery for the Kering group unifying customers across all luxury houses (Gucci, Saint Laurent, Balenciaga, Bottega Veneta, etc.) via crossbrand identity resolution on top of purchase-history fact tables. • Consumed by CRM and marketing teams across brands for activation and by group-level analytics / BI for executive reporting Led the maintenance of the legacy AWS data lake and the migration of core Spark / Scala pipelines (5-15 critical jobs) toward GCP / BigQuery without business disruption. • Contributed to the team's data mesh patterns (DBT project templates, domain ownership conventions, testing standards) adopted by other engineers on the customer domain. Stack: BigQuery · DBT · SQL · AWS (S3, EMR) · Spark · Scala · Python · Data Mesh · GDPR

• Stepped in as interim Lead Data Engineer during a team transition

• Optimisation du BigQuery Storage et de l’usage des Slots : -25% sur les coûts BQ et -30% sur le runtime des pipelines business critiques • Refonte des core ETL pipelines via Airflow et BigQuery • Ajout d’un outil de monitoring basé sur les BQ audit logs

Skills: Airflow, PostgreSQL, Django, Python, SQL, Metabase, Pandas, Jupyter • As the only data engineer, led the design of the data warehouse and ETL integration of multiple data sources using Airflow and Pandas. • Actors deduplication using classification ML model • Developed quality checks tooling using external Open Data APIs like Annuaire Entreprise and BAN

• Designed and scaled IEBS (Impression Economy Balancing System) — a multi-objective, constrained re-ranking system plugged into page construction across key surfaces (podcast, audiobooks). Balances financial goals, strategic content targets, and user-centric engagement guardrails via pacing, boosting, and content-identification heuristics exposed as gRPC services. • Led the design and implementation of financial utilities for recsys — translated royalty economics and strategic priorities into constrained-optimization signals consumed by ranking layers. Drove $10M+/year in royalty savings and activation of new verticals (audiobooks, video podcasts). • Built offline evaluation tooling (Ray + Flyte + gRPC) enabling more efficient experiments on home. • Led the migration of royalty-calculation ETL pipelines from Luigi to Flyte, unlocking companywide compliance and further platform development. • Leadership: drove design reviews, mentored engineers, collaborated cross-functionally with finance, content strategy, and data platform teams; presented results to leadership. Stack: GCP · Dataflow · BigTable · BigQuery · gRPC · Scala · Java · Python · Ray Flyte · SQL

Skills: GCP, Airflow, Dataflow, BigQuery, SQL, Java, Python, PowerBI, Tableau, Druid • Built video-categorization ML pipelines (training, serving, evaluation) aligned with IAB content taxonomy. • Revamped the partner analytics platform by migrating from Exasol + HBase to Apache Druid → $200k/year licensing savings. • Designed partner-facing data models tailored to concrete business use cases.

Skills: CDP, Talend, Kafka, HDFS, Hive, Warp10, SQL, Java, Python, Pandas, HighCharts, DataViz • First Data Engineer, 7th out of 100 employees. • Deployed Cloudera Data Platform on public cloud OVH and built a data warehouse for energy consumption data. Hadoop, HDFS, Hive • Developed REST APIs and data pipelines using Python, NIFI, Spark, and Vertx.