Problem
Large enterprises applying for grants, especially in innovation and green energy, often face a highly complex application-writing process. It requires not only domain expertise, but also the ability to define correct objectives that match the call’s requirements, and to craft a convincing narrative built around the right keywords to maximize the chances of funding.
A major difficulty is also understanding what the application actually requires and assessing how well the company meets those requirements.
Challenge
A key constraint was the requirement to use open-source LLMs, which typically have more limited capabilities than proprietary models, both in answer quality and in context window size. In this case, large context was critical, as grant documents can exceed 100 pages.
We addressed these limitations with an advanced agent-based architecture and by splitting the workflow into multiple steps, including keyword generation, executive summary generation, SMART objective creation, and a Peer Reviewer module. The Peer Reviewer used a different LLM to evaluate earlier outputs and ensure alignment with the application requirements.
While it was easy to find plenty of material from grant calls, a major challenge was the lack of strong reference applications from the specific domain we were working in. We conducted extensive research, but most publicly available examples were grant applications for academic research rather than industry-focused initiatives.
Solution
We built an AI assistant for writing grant applications using an agentic approach, leveraging the Langflow framework. The system integrates large language models (LLMs) to support both efficient text generation and processing of complex grant documentation. We used open-source models to enable on-premise deployment for enterprise clients, providing stronger guarantees of confidentiality and data security.
The assistant consists of several capabilities, including keyword generation, drafting the application objectives, and producing an executive summary that can then be expanded into subsequent sections.
A simple user interface was created to enable rapid deployment and testing during the Proof of Concept (PoC) phase. The system runs on AWS Bedrock, enabling efficient model management and optimization for cost and performance.
The result is a modular, scalable tool that automates key stages of grant-application drafting, using modern AI technologies and allowing easy integration with other systems.
Result
During the PoC phase, the client’s grant-writing experts evaluated the tool very positively. The most valuable feature was keyword and key-phrase generation. We estimate that such an assistant can reduce application-writing time by up to 50%, while also increasing the chances of securing funding through better-defined, requirement-aligned objectives and a more convincing narrative.
