Arthur Lauga

Launch of a Data Lake at Tiime: strategic project management, establishment of the roadmap and priorities. Full internalization of Change Data Capture (CDC) flows into industrialization via Flink+Kafka; data exposure on S3+Databricks

Improvement of the OCR document processing pipeline: benchmark of new AI / LLM solutions, cost-benefit/risk analyses. Expenses divided by 3 following the different iterations.

Generalization of the matching algorithm to all types of documents processed by Tiime; >90% of the target processed automatically, compared to 50% before.

• Design and deployment of a hybrid prediction algorithm, producing accounting categorization and detecting the third party in the bank statement:

- 30% reduction in error rate

- Enabled Tiime to significantly diversify its target accounting firms by offering a much more flexible solution. Approximately 150k additional managed files (vs 40k before)

• Design and production deployment of an invoice matching algorithm (Tiime Invoice) with the corresponding payment transaction. Automates >80% of concerned invoices.

• Work on the stability of Deep Learning models on a changing production population

• Modeling of bank transaction annotation (account number, VAT):

- Deep Learning, Random Forest, and Nearest Neighbor in a unified pipeline

- Very noisy training data from heterogeneous sources

- Close collaboration on feature engineering with business experts

• Development of an unsupervised recurring payment detection algorithm

• Work on model robustness in production via Monte-Carlo Sampling. Modeling of bank transaction annotation (account number, VAT): - Deep Learning, Random Forest, and Nearest Neighbor in a unified pipeline - Very noisy training data from heterogeneous sources - Close collaboration on feature engineering with business experts • Development of an unsupervised recurring payment detection algorithm • Work on model robustness in production via Monte-Carlo Sampling.