Benjamin LAZARD

Co-creating a data department from scratch. 1st mission: an API to help CTI analysts detect relevant fields for a STIX graph (malwares, attackers, etc.) - Set up team standards for collaboration linting, unit tests, documentation, CI/CD, coding env, etc. (black, mypy, pytest, Drone, notion, git flow, vscode settings) - Built a light ETL: scraping & querying, standardizing jsons/xmls, building CLI tools, etc. (python, scrapy, pycti, JMESpath, minIO, argparse) - Defined evaluation metrics and a training procedure to compare different NERs (0‑shot, fine‑tunable) (GliNER, Docker, Open‑CTI)

Forecast Demand @Turbo darkstores 2nd Datascientist in the team - Create the 1st local production pipeline of the team converting notebooks into a git-based repo with modular code (python, SQL). - Design a DAG (Kedro, Airflow) to scale local pipeline to remote execution - Create the 1st training procedure for a timeseries model (sklearn, custom classes), ensuring compatibility with various frameworks, building a common evaluation procedure for the team while adapting to our own constraints. - Implementing industry models like FB prophets and developing our owns (sklearn) => Operating in all LATAM We are adapting existing timeseries models to our needs (forecasting for several products at once, evaluating demand while accessing only sales, out of stock, new products, growth related to exogenous data we don't have access to, etc.) Recently started product recommendation and bundling Everything is to be done yet ;)

Data for better decision making in a FoodTech - Create data flows (Snowflake, AWS-lambda, DBT - python, sql) to gather orders, employee shifts, French population geostatistics, suppliers APIs, etc. - Elaborate visualizations (Looker) - Find partners and competitors through scraping, processing and unification of various websites (Scrapy). - Build from the ground up a custom map-based web app (docker, AWS-ECS, leaflet, flask - javascript,hml,css,python) to display our competitors and partners in an interactive fashion. the ability to check their ratings, location, cuisine, availability on delivery platforms, neighborhood wealth, etc.

Revenue model - Churn Prediction Youdrive team: connected car insurance (using telematic data to compute a customized premium for each customer). - Create a dedicated python package for churn prediction (pyspark, pandas, git, MySQL) - Improve dashboards to monitor this new customer segment - Created a dedicated python package for churn prediction (pyspark, pandas, git, MySQL) - Improved dashboards to monitor this new customer segment

Design a new offer for online retail - Review former and new offers compatible for online sales, find and estimate target market size - Elaborate a price estimation procedure to limit friction, and insure compatibility with logistics - Suggest various solutions for the implementation of a new website (studying scalability, price, convenience, compatibility with inventory, etc.) ‑ Business plan New website live.

[Japan] Designing machine learning algorithms to enable robots to cook, together with Carnegie Mellon University (CMU). - Implement algorithms for recipe recognition/generation - Design new deep learning architectures - Study state-of-the-art research papers (less than 1 year old)

[Spain] Full-stack dev. for bookdepository.com division. - Create and deploy query-specific advertising widgets, integrating with various systems - Update Caching policy using AWS tools (partial design + implementation). I was extended an offer.

Volunteering in a newly created food startup. - Code a python script to automatically generate PDF business reports with charts monitoring KPI based on the cash register´s spreadsheet output (python, Latex). - Scalability in mind (easy to adapt from spreadsheet input to database input, possibility to add new charts, and dynamic html plots) Used to showcase the company's performance to investors.

[UK] Facilitating the paint-shop in JLR Halewood plant techiness. - Collect and Display data aimed at controlling the quality of a process. - Monitor improvement works executed by contractors. - Design and code a webpage to help paint-shop engineers use the different IT services. -Redaction of SQL queries aimed at detecting cars stuck in the line.

Solicit Companies to get them to participate to the forum (5000 visitors, 100+ companies). Most efficient team of 3 (number of companies recruited) @ Telecom Paristech.