Neste: Revolutionizing renewables innovation with AI-powered knowledge discovery
Neste embarks on a new journey, using Generative AI in innovation and technology development to augment information discovery and creation processes with a new AI-powered knowledge assistant solution.
The challenge
Neste, a global leader in renewable and sustainable aviation fuel (SAF) and renewable diesel, is fundamentally transforming from a traditional oil refiner into a provider of sustainable solutions. This shift requires extensive innovation as the company explores future sustainable raw materials, refining processes, technologies, and new business models to support a low-carbon future.
Neste’s Technology and Innovation unit, consisting of leading experts in chemical research, technology, and engineering, plays a crucial role in the company’s transformation journey. Finding precise information quickly and efficiently from a massive, complex, and scattered knowledge base has always been a challenge for experts.
Furthermore, leveraging such complex and scattered information for decision-making and new knowledge creation would require further manual synthesis and consolidation of information.
To improve this time-to-information challenge, Neste Innovation and Technology Digitalization team partnered with Futurice and the Futurice Family company Recordly to develop an AI-powered research assistant. The goal was not just to improve search but to create a conversational experience that enables experts to find relevant and contextualized answers to their questions and create new knowledge for future use.
What we did
Futurice and Recordly collaborated with Neste to design and build a Retrieval-Augmented Generation (RAG) solution with a conversational user interface. Experts can initiate and maintain multiple natural-language conversations and receive precise answers to their questions, enriched with references to the original documents.
To ensure accurate, efficient, and scalable knowledge retrieval, we designed a state-of-the-art hybrid search engine using semantic and BM25 search techniques that searches across a vast and well-curated corpus of research literature. The search uses agentic intelligence to improve search queries to find the most relevant information, helping experts to discover trustworthy, high-quality results.
To further enhance response accuracy and relevance, we integrated contextual ranking, a cutting-edge technique from Anthropic on top of the standard RAG approach.This improved the system's ability to surface the most relevant insights.
Using our use case discovery and design framework, we were able to actively engage the end users in shaping the solution and implementing a robust evaluation framework to measure the relevancy, correctness, and faithfulness of AI-generated responses.
To bring this all together, we designed an intuitive conversational user experience that encourages dialogue between the user and the solution, maintaining the history and contextual memory of the discussions across topics.
Why it matters
By leveraging AI to automate knowledge discovery, Neste can improve efficiency and productivity in Innovation and Technology Development. The solution accelerates decision-making, enhances collaboration, and helps Neste maintain its competitive edge in renewable technology.
As part of the solution, we also established a state-of-the-art Gen AI foundation architecture consisting of highly abstracted, modular, and scalable components in terms of a re-usable design and code blueprint, opening up further opportunities for Intelligent Knowledge Agents solutions across Neste.
Besides accuracy, explainability and transparency are also of paramount importance for the use of AI in research due to the critical impact on decision-making. The above-mentioned modular architecture and evaluation framework enabled us to experiment with a variety of parameters and language models in rapid iterations until we achieved measurably high benchmarks on these metrics.
The solution not only expedites time to information but enables us to be better experts.
As the solution introduces a new way of discovering and creating knowledge for Neste, we designed an adoption process involving continuous learning and integration with the existing processes and ways of working to maximize the value creation.
Finally, the established robust architectural foundation and learning and process integration experience will enable us to evolve and scale the solution to meet constantly changing business needs and the rapidly developing Gen AI landscape, paving the path for a broader AI-assisted future at Neste.
As AI continues to fundamentally transform scientific research, Neste is positioning itself at the forefront of this evolution through this first milestone in its journey towards a Human-AI collaborative innovation and technology development.
Technologies used
- LlamaIndex
- RAGAS
- Vector Database
- BM25
- Hybrid Search
- Azure OpenAI
- Google Vertex AI
- Azure Document Intelligence
- Reranker
About the client
Neste (NESTE, Nasdaq Helsinki) creates solutions for mitigating climate change and accelerating a shift to a circular economy. The company is the world’s leading producer of sustainable aviation fuel (SAF) and renewable diesel, enabling its customers to reduce their greenhouse gas emissions. Neste refines waste, residues and other renewable raw materials to high-quality renewable fuels at its refineries located on three continents. The company’s annual renewable fuels production capacity will be increased to 6.8 million tons in 2027. Neste has high standards for sustainability and the company has consistently been recognized by several leading sustainability indices. In 2024, Neste's revenue stood at EUR 20.6 billion.
Talk to our experts
Get in touch
Looking for help with an idea, brand new brief or in-flight project? Drop us a line for a straightforward conversation.