Matteo Peluso@mozzarellaman

There are currently no software solutions for reconciliation that do not require a constantly updated digital record or that, in other words, allow ‘one-offs’. Accountants serving large limited companies, that take advantage of recurring bookkeeping have no trouble when it comes to tax season. But, the 30 million sole proprietorships in the US and UK, that do not have access to these luxuries of scale, are forced to rely on, more often than not, the manual, laborious exercise of reconciliation. This problem is exacerbated when considering the fact that, in most cases, invoices and receipts are stored in non-digitized formats and require accountants to handle them as ‘one-offs’. This occurs as sole-traders typically do not maintain digital books or a track record of any kind. Enter stage right: ReconcilAi. Our app allows accountants to upload bank statements, invoices, receipts and contextual information. Natural language processing is used to process documents on both sides of the equation, standardizing the format and utilizing contextual understanding to retrieve the appropriate values. After which, simple algorithms undergo the matching process and present the results in a user friendly manner. The user can then download their data and proceed confidently with their tax-return process. Even in the case where only 50% of documents are matched successfully, the user work load has been halved saving countless man-hours and wages. Building this application was largely simplified thanks to tools such as, but not limited to, Trulens for evaluating our chosen model’s contextual retrieval abilities and Portkey for rapid prompt prototyping in practice. We would also like to thank the LablabAi team for organizing the insightful workshops and the hackathon as a whole.