Javier Garcia de Leaniz

NLP models that I trained and developed to identify, extract, and structure data from veterinary invoices to allow for the reimbursement of animal health insurance claims.

I developed a pipeline consisting of OCR, layout detection techniques, and named-entity recognition (NER) models to extract the relevant information from the invoices accurately. I also built the validation module to identify and validate medical diagnoses against the policyholder coverage.
Finally, I developed the extraction confidence methodology to help determine claims reimbursements to be processed automatically or reviewed by a human, depending on the different models' confidence and business rules.

NLP models that I trained and developed to identify key data points from mortgage contracts to allow automatic audit and validation of data in actual contracts vs the ERP.

I trained a model to classify between main contracts and their annexes, extensions, and modifications using TF-IDF features to train a classifier. I also developed the validation module to disambiguate and match contracts and DB rows and perform the comparison to highlight differences.

NLP models that I built and trained to extract key data points from various documents such as invoices, mortgage payments, paychecks, and more.

The goal was to increase the efficiency of the application process for a tax relief program offered by the government due to COVID-19 that received millions of requests.

I developed a pipeline consisting of handwritten text detection, layout detection techniques, classification (to detect the document type), and NER (named-entity recognition) models to extract the relevant information from the documents accurately. I also developed the confidence module that prioritized manual review of applications based on business rules and models' confidence.

I developed an AI-driven system to automate the analysis of financial documents in construction processes, targeting the high volume of invoices, purchase orders, and goods received notes. This project aimed to detect mismatches and inconsistencies to prevent financial losses.

The project's pipeline started with OCR technology to extract text from scanned documents accurately. I employed named-entity recognition (NER) models to identify and categorize key data points within these texts, such as vendor names, dates, and amounts. An important part of the project was the development of classification models to accurately detect and categorize different document types, automatically detecting its relevant data points.

Additionally, I implemented fuzzy matching algorithms to link items listed in invoices with corresponding entries in purchase orders. This approach was key in identifying mismatches and inconsistencies.