Overview
This Librios implementation reduces researchers' time by c.20% by enabling AI-powered initial drafts of company and individual due diligence reports. Librios stores historical reports in their database, aggregates news articles from the web, and scans watch lists with government KYC and KYI data sources.
The information is processed into granular sections, textual analysis links the context and relevancy then using RAG (retrieval augmented generation) AI prompts build the GenBI insights as draft reports.
Building confidence within the research team that the GenBI outputs will form the foundation of their confidential technical reports was the primary challenge.
Objectives
•Improve the speed of gathering DD/Screening intelligence from multiple sources
•Use organizational knowledge to enrich the reports based on 1000’s of historical reports
•Aggregate adverse news articles around human rights due diligence information
•Export into a MS Word document for manual curation, then upload final report back into the service for future research enrichment.
Outcomes
Predicting an aligned outcome
Using conformal prediction, the statistical technique used to make predictions with a measure of confidence. We provided a way to construct prediction intervals or sets that have a guaranteed coverage probability, meaning that the true outcome will fall within these intervals or sets a specified proportion of the time. We breakdown this as:
•Training Phase: we start by uploading corrected DD reports and the Librios platform detects the most relevant data points to inform the AI to generate recommendations and predictions.
•Calibration Phase: after training, we use a separate set of data reviewed by the senior technical reviewers (called the calibration set) to measure the model's prediction errors. We calculate how far off the model's predictions are from the actual values.
•Prediction Phase: for a new data point, the conformal prediction method uses the information from the calibration phase to create a prediction interval or set.
This interval or set is chosen so that it will contain the true value with a high probability (e.g., 92%).
The key advantage of using conformal prediction is that it provides a principled way to quantify the uncertainty of predictions. It can be applied to most predictive models and works with minimal assumptions about the data distribution.
Outcomes of Data Driven Interventions
The insights and analysis driven from the data gathered in the service from the uploaded documents and aggregated data provides an opportunity to assess new information points. For example, extended data could deliver researcher input strengths and weaknesses mapped against interventions such as standardization, refresher training for the team, recommended technical reads, mentoring or other culture training exercises.
We respect our client’s confidentiality as this is giving them a competitive advantage.
Description: 