AI-powered data extraction from expert reports on waste streams

The European Union generates approximately 2.2 billion metric tons of waste annually (Eurostat, reference year 2022). This waste volume essentially represents a significant source of secondary raw materials, the use of which, however, has so far been limited.

To be able to reintroduce waste into material cycles, it is possible to confirm that a material has reached the end-of-waste status by demonstrating certain material properties through technical and scientific expert reports. This means that such evidence verifies the transition from waste to a secondary raw material.

In practice, these expert reports, referred to as “assessment records”, are available at the Federal Ministry of Agriculture and Forestry, Climate and Environmental Protection, Regions, and Water Management as PDF documents and contain both continuous text and tabular analytical data, images, and handwritten notes. This heterogeneous structure, combined with the annual volume of documents, significantly complicates systematic digital evaluation for the purposes of waste and quality management. With the increasing volume of documents, it is no longer feasible for case workers to manually review all records in their entirety.

Existing assessment reports are not standardized and therefore have a highly heterogeneous data structure. Their scope, layout, structure, and level of detail vary greatly depending on the assessor and the waste stream. Technical terminology can also vary depending on the industry, region, and assessor. Scanned documents, incomplete tables, photos, or handwritten additions make automated analysis difficult. Depending on the assessor, these documents are submitted via the Federal Ministry’s existing database (the Electronic Data Management - Environment (EDM)) or by email, although the latter method of submission is gradually declining. In the future, submissions will only be accepted via the database.

Since waste classification regulations for additional substance groups are currently being drafted, the number of expert reports is expected to grow in the future. To make the review process more efficient and improve data extraction and analysis, the Federal Ministry is now seeking an AI tool to automate this process.

We are looking for an AI-powered tool that automatically parses submitted end-of-waste assessment reports and converts the information they contain into a structured, machine-processable, and machine-analyzable format. This includes extracting, validating, and analyzing data from heterogeneous PDF documents containing text, tables, and images. In addition, the tool should be able to capture handwritten annotations. Optionally, images can be extracted from the assessment reports, stored, and linked to the corresponding page in the original report, allowing a reviewer to quickly navigate through the assessment report.

The tool being sought should classify uploaded documents and compare extracted values with existing thresholds. Any anomalies should be reported to and displayed for the reviewer. A fully automated assessment of the uploaded documents is not intended. The substantive evaluation and interpretation of the results are intentionally left to the reviewer.

The structured storage of data is intended to increase the level of standardization in the expert reports. In addition, submitted assessment documentation will be reviewed for completeness and compliance with applicable regulations. Optionally, the tool can automatically verify whether the correct documents have been uploaded to the correct section in the Federal Ministry’s database (Electronic Data Management - Environment (EDM)) to expedite review processes.

The extracted data will be processed in such a way that it can be used to further evaluate the current condition of secondary materials. The goal is to assess developments in a historical context and to provide an overview of contaminant levels and resource potential as a basis for planning and regulation. Extracted data should be available for download as a PDF and in a standardized file format. Based on this, it will finally be possible to establish reliable and traceable quality standards for secondary materials. This data provides an important foundation for policy-making and the optimization of the circular economy..

The following core functions should be covered in any case:
• Information extraction: Automatic identification of relevant content (threshold values, material designations, analysis results, waste codes)
• Table and data extraction: Extraction of measured values and parameters from structured and unstructured table formats
• Document classification: Assignment to waste type, treatment technology, and legal scope
• Compliance check: Automated comparison of extracted values with the applicable limit values of the relevant regulation
• Summary: Creation of concise summaries for case handlers

We are seeking tools that utilize EU-compliant AI models and are compliant with current standards such as the EU AI Act and the GDPR. To avoid having to set up additional infrastructure at the Federal Ministry initially, we are looking for solutions that can be operated on a dedicated instance in a sovereign cloud and, if necessary, transferred to their own servers. In addition, interfaces to data warehouses and other systems, as well as access rights management, are required. Both project management and documentation must be conducted in German.

Non-goals:
• Standardization of expert reports through template forms for experts
• Establishment of external repositories
• Establishment of new infrastructure
• Tools that distort values or generate false results

Take this opportunity to present your innovative approaches. Companies, startups, and research institutions are invited to submit their proposed solutions. Please provide a brief description of the solution, the costs (including both procurement and maintenance costs), and a proposed project timeline.

When estimating costs, base your calculations on a growing number of expert reports. Assume 200 reports per year for the first year, increasing to approximately 2,000 reports per year over the next 2–3 years. Depending on the subject matter and type of report, the length can range from 20 to 400 pages.

Description:
Submit your idea using the online form. Make sure your concept meets the requirements and evaluation criteria. Make a strong case with clear use cases and references.
You’ll also need a cover image. This will serve as your visual representation in the overview of submissions for this challenge.
Optionally, you can also attach a file. This file should supplement—not replace or repeat—the text in the form fields! Use the file attachment, for example, for graphics or presentations.

Summary of Added Value:
Summarize your unique selling point and key advantages! What makes your solution particularly innovative? What sets it apart?
Use cases, references, and a realistic project timeline will help with the evaluation.

Confidential Information (for the jury only):
This information can be submitted confidentially via the platform.
You can enter additional information in this text field that you wish to share exclusively with the project managers. Alternatively, you may send a PDF via email to the moderators of the challenge.

IMPORTANT NOTICE:
Submissions in consortium are possible. Please keep it brief (guideline: max 10-12 presentation slides or 4-5 DIN A4 pages). This challenge is part of a market exploration, so it is not necessary to prepare drafts or feasibility studies specifically for this occasion.