Christophe Duong

– Working as a freelance helping companies scale their AI infrastructure and practices around data engineering, analytics and data science.

– As part of a fresh startup, DinMo is building an AI-powered platform to help businesses get started with a composable data platform (CDP) that simplifies data integration, management, and analytics. – My role as an early data architect on the team, involves designing and implementing the data infrastructure from scratch. Thus, I selected an orchestration tool (Dagster) as the backbone of the data platform and started to put in place a whole framework versioned in python to easily create, scale and maintain a large collection of data connectors (reverse-ETL) for data integration between major data warehouses with various CRM, marketing tools, and databases for data activations. – I also created all the necessary foundational code abstractions in Python to manage different workloads to perform data transformations via dbt, monitoring, alerting, and logging of data pipelines that handles either model predictions, or advanced uses cases such as A/B tests experiments, identity resolution, and semantic layers for analytical metrics computations.

– As a senior technical lead on one of the data team at Sodexo, I was responsible for leading the design, coordinating implementation of data pipelines and models to support the "product-swap" recommendation engine. – I collaborated with a data product owner, while managing a team composed of a data scientist, a web-developper and 2 data engineers. – This engine deployed on Microsoft Azure data platform and Databricks leveraged machine learning algorithms (LLMs) to perform text analysis and natural language processing (NLP), and provide product recommendations to optimize supply chain management for Sodexo's worldwide line of restaurants and catering services. – My responsibilities included overseeing the end-to-end data workflow, from data ingestion and preprocessing to model training and deployment on a web application serving as the user interface for Sodexo's internal teams, facing challenges related to data quality due to international contexts (multiple languages), scalability (multiple countries with large amount of transactions), and real-time performances (trade-offs versus offline processing).

– Contributing to the early engineering team at a startup, we first built an open-source ingestion tool written in Java and Python to handle a large portfolio of APIs and data warehouses (Snowflake, BigQuery, AWS, Redshift, Databricks, etc). – I helped design and develop the ingestion product that would be key to any data teams, reaching the Product-Market-Fit, growing the project and community from less than 100 to 10k GitHub star which is now being widely used worldwide. – To further drive this development, I also bootstrapped the initial internal data infrastructure, hired and managed a team of Data Engineers and Data Analysts to enable deliveries of analytical insights or Business Intelligence with self-service capabilities to other teams focused on the business operations (finance/support/marketing), engineering, nurturing cross-functional collaborations using the Modern Data Stack and heavily leveraging open-source projects such as dbt, sql, dagster and metabase.

– Starting from zero, I built the data infrastructure using Python, dbt, sql and Looker that powers the matching engine (elastic search & recommendation engines) that analyzes and automatically connects freelancer profiles with missions from corporate enterprises. – After setting up the technical foundations of the data warehouse on GCP/BigQuery and ETL pipelines in Airflow, I recruited, hired and managed a team (composed of Data Analysts, Data Engineers and a Data Scientist) that I was leading to coordinate and collaborate with product development and business teams through Business Intelligence and Data Science applications.

– Worked as a bridge between the Data Engineering team and Data Scientists to transition a recommender system from prototype to production state while scaling it up to support increasing number of visitors, recommended sales, business units and business applications. – Other activities included mentoring A/B testing methodologies or supporting other departments (Business Intelligence, Marketing, User Experience, Operations, HR, etc) with occasional ETL jobs, data analysis, modeling, etc.

– Designing and building large scale recommendation system, classification, clustering models with machine learning pipelines on Amazon AWS Elastic Map Reduce (EMR) using Hadoop/Spark in Scala

– In parallel to studying a Master degree at IE Business School, worked part-time at a consulting firm specialized in statistical modelling using Bayesian approach. – Advised, coached and collaborated with consultants professionals knowledgeable in Statistics, Mathematics, Economics and Technology on latest trends around Big Data advances. – Discussed, planned and communicated how more traditional, advanced statistical methods such as Bayesian Time Series Forecasting could be integrated and improved in a Big Data context such as Apache Spark with real-time stream analytics.

– Designed and developed both batch and real-time processing pipelines (similar to Lambda Architectures) as part of the maintenance efforts of a global financial software platform deployed in production in several financial institutions (banks, hedge funds etc) – Contributions included programming in Java or refactoring distributed and multithreaded systems while assisting Service Desk teams to resolve SLA from clients, coordinating with DevOps teams on product releases, coaching peers on distributed version control best practices, collaborating with QA teams to enrich unit, integration and acceptance tests in Groovy and managing migration of database schemas with DBA teams with SQL scripts; which led me to cover a wide range of aspects of a mature software lifecycle.

– Prospected and researched new innovative opportunities as an "intrapreneur", interacting with customers to reduce business and clinical uncertainties of our ideas while building C++ prototypes with heavy usage of GPU rendering (parallel processor for real-time applications), integration of machine learning and image processing tools to track, segment and analyze physiological functions from 4D Ultrasound images. – Day-to-day collaboration with a highly diverse team: Clinical Personnel, Mechanical Engineer etc – Even assisted physicians perform a minimally invasive heart surgery on a live patient during clinical trials by driving and demonstrating one of the prototypes that I created to help improve patient outcomes.

– Mentored by computer vision researchers, I integrated image processing and machine learning algorithms together with 3D visualization rendering tools during a competition of the "Deep Bleeder Acoustic Coagulation" program funded by DARPA (an agency of the US Department of Defense), developing the user front-end of a C++ prototype to automatically detect and localize internal bleeding of limbs and trigger ultrasound beams therapy on the battlefield. – As the winning team of the federal grant, DARPA selected Siemens to exclusively pursue the second phase of the program. – The goal is "to build a portable device that uses focused sound waves to treat troops bleeding internally from wounds sustained on the battlefield. Designed so that soldiers can operate it with minimal training."

– After successfully completing a course on the implementation of a compiler in C++ using modern object oriented programming techniques, i was selected to co-lead a team of 50 assistants to teach programming techniques and languages to 300 students, planning the topics learned during a whole year through projects then coaching students to accomplish them: "The best way to learn is to teach" https://www.lrde.epita.fr/~tiger/assignments.html#Tiger-2008