Our team is part of the company Genie Enterprise Inc., which is a member of the Ecoliance Rheinland-Pfalz, Germany. We feel responsible for our environment and support projects that focus on sustainability and achieving the UN climate goals.

Our AI company specializes in the digitization of human perception and knowledge. In particular, the field of natural language processing is one of our core competencies. So the "Climate Policy Radar" challenge is a great candidate that we can support with our expertise.

What it does

So many documents and texts have been provided by the challenge setter (1.6GB of text). It's hard to find answers to different questions in the huge amount of data. We think it is helpful to have a tool where you can ask questions in normal human language and the system searches for the answers to your questions. So we built a PoC version of such a system this week.

How we built it

We have designed and built a system, that takes the documents provided by the challenge setter, their metadata and ingests them into a special search engine based on ElasticSearch. During the ingestion process we applied some preprocessing to extract text from documents, to clean, to normalize and to standardize the given text data. BERT algorithms are used to understand questions in human language, search document candidates in the search index and answer the question from the identified documents. A simple web interface is provided to interact with the system.

Challenges we ran into

We are using some of our proprietary tools in the preprocessing of documents and data. As these tools are used by and developed together with our clients, we can't open source our solution in this hackathon scope. But we would love to support "Climate Policy Radar" in its further development.

Accomplishments that we're proud of

Building a working solution in a short period of time.

What's next for Genie Policy Reader

We will continue working on this toolchain to make it part of our solutions portfolio.

Built With

Share this project: