Firas Omrane Email & Phone Number

Paris, Fr

New York, Ny, Us

New York, Ny, Us

IAS processes over 1T impressions/year.My MLOps team classifies multi media content to help advertisers make more suitable and safe placements of their ads, by helping either target or avoid bidding next to social media content.We handle on average ~50M multi media content (videos, images, text) / day coming from Ads platforms (facebook, instagram, tiktok, youtube ..)To classify a content we use multiple ML models that detect raw signals (logos, scenes ..), then aggregate these signals and apply rules made by our product team to trigger the classification.The processing is done in streaming and results are sent to downstream consumers and to a data warehouse for monitoring, dashboarding and quality assessment (sampling, annotation, models quality assessment)- Led the architecture and implementation of our real-time data lakehouse, on top of Databricks to enable monitoring and models quality assessment. Adopting a One Big Table modeling (denormalization), liquid clustering, partitioning and pre-aggregations to facilitate fast dashboards and optimize reads.- Redesigned and implemented our stateful classification service to be more fault-tolerant and saving ~50% of cost leveraging RocksDB as state store.- Implemented dynamic scaling based on demand of our ML inference capabilities, leveraging Kubernetes and KEDA for event-driven scaling to maintain SLAs at peak times.- Enhanced our grafana-based monitoring of processing rates and queues consumption for real-time system health assessment, and CI/CD workflows.- Refactored and optimized a big part of our codebase, improving design and interfaces, reducing bugs and increasing test coverage.- Incidents debugging and management, Interviewing..Tech stack:Python, Docker, Kubernetes, Helm, Keda, RabbitMq, AWS (EC2, SQS, S3, ECR..), FastApi, AsyncIO, Spark streaming, Databricks, Terraform, Grafana, Jenkins (CI/CD).

Zurich, Zurich, Ch

Working on providing best in market real estate data to our clients:At pricehubble we help provide the best data insights to our clients and make their job easier.The data is core to our products, which is consumed directly or via our ML valuation models.- Defining and contributing to our technical roadmap to build and consolidate our data platform that supports: Data quality, Monitoring, Alerting, Versioning, Schema changes, Data governance, Scaling, Infrastructure.- Migrating and renovating our infrastructure with cloud costs in mind: Architecture and design, Iaac, Spark on k8s, Permissions.- Built a queries performance benchmarking tool that helped decrease DB queries time and find best indexes.- Enable more frequent retraining of our valuation models on fresh data, having data consumers with different needs of our ML deduplicated data: One optimizing for recall so customers have always access to all the data points, and data fed to train valuation models optimizing for precision to prevent duplicated data and creating overfitting.- Make sure our systems scale with the growing processed volumes of data (> 1B images and data points) and respect SLAs: Built an experimental queries performance tool that helped decrease DB queries time and find best indexes.- Collaborate with stakeholders to respond to their data requests by building pipelines ingesting data from different sources (Bigquery, APIs, GCS) and exposing them via REST API, Relational DB. Parquet.-Helped migrating our data acquisition system to Redis as a key value store, to speed it up and save computational time.- Led the data side of entering 2 new countries.- Contributing in: Planning, Prioritization, Requirements specification, Project management, mentoring, interviewing, knowledge sharing, refactoring, respond to customers questions, bug fixing and documentation.Tech stack: Python, Docker, Kubernetes, Spark, SQL, Airflow, Postgres, GCP, BigQuery, CI|CD, Redis, Terraform, FastAPI, Luigi

Zurich, Zurich, Ch

Grocery Delivery Service Start-up (www.livri.tn )- Developed functionalities and interfaces for both client and administration websites.- Built and tested REST API endpoints.

Tunis, Tn

- Maintained signal and network for clients.- Communicated with clients to solve their problems.Leveraged knowledge in Telecommunications networks.

Engineering Degree, Mathematics And Computer Science

Master Of Science - Ms Data Science

Telecommunication Engineering

Full Stack Web Development Certification, Computer Software Engineering