Optimizing Nonprofit Grants with Generative AI
Optimizing Grants with Generative AI
Customer confidential
Harnessing the power of generative AI, this non-profit customer teamed up with SDG to enhance its grant analysis process.
Challenges
Latency in data retrieval and analysis
Understanding and implementing the generative AI Stack
Disparate data formats and sources
Improving grant evaluation
Our customer is a national nonprofit dedicated to improving access to higher education and career success among underserved populations. Through strategic grants and program-related investments, their foundation supports innovative solutions that increase college success and career readiness.
Since May of 2015, this foundation has invested over $300 million in grants to nonprofit organizations that expand educational opportunities and foster upward mobility for underserved communities. To enhance the speed and accuracy of their grant evaluation process, they partnered with Solution Design Group (SDG) to explore using generative AI to identify patterns and provide actionable insights in real-time. By addressing challenges like latency and data relevancy, SDG and the customer laid the groundwork for a responsive, AI-enhanced grant analysis workflow.
Streamlining grant analysis
This foundation’s commitment to expanding educational access is supported by its rigorous grant evaluation process, which relies on deep analysis to make high-impact funding decisions. With over $300 million in grants awarded, the foundation faced increasing demands for accurate, real-time insights to guide its investments. Recognizing the potential of generative AI and Retrieval Augmented Generation (RAG) to meet these needs, they collaborated with SDG to prototype a comprehensive generative AI technology stack that would accelerate the grant analysis process while ensuring that recommendations were data-driven and contextually relevant.
With a wealth of historical grant data and real-time application information, the foundation was well-positioned to leverage RAG-powered LangChain agents to process and analyze data efficiently. SDG’s implementation allowed them to go beyond conventional analysis, gaining real-time insights and retrieving information directly related to each grant’s objectives, applicant history, and outcomes.
The right approach
Leveraging the full generative AI stack
SDG’s consultants recommended that the customer use a generative AI technology stack, comprising AWS, LLMs, and LangChain agents. This approach included AI models that retrieved relevant information, generated summaries, and offered actionable insights tailored to each grant’s criteria. SDG worked with them to orchestrate a collection of LangChain and LangGraph agents to rapidly analyze incoming grant and demographics data, reducing the need for manual data processing and allowing analysts to focus on high-impact evaluation and decision-making.
- Real-Time Data Retrieval and Adaptive Analysis
LangChain and LangGraph automated agents could quickly retrieve and analyze information from AWS data sources using Retrieval Augmented Generation (RAG), providing the foundation’s grant analysts with immediate access to applicant insights and grant impact metrics. - Dynamic Recommendation Engine
LangChain Agents used generative AI to identify potential risks and assess a program’s alignment with the foundation’s mission, ensuring each decision was backed by robust, real-time data.
A new standard in grant analysis
A significant challenge the team faced was orchestrating the integration between different tools to create an efficient pipeline while minimizing latency and hallucinations. Additionally, the complexity of retrieving real-time data through RAG while maintaining conversational coherence presented unique challenges for engineering prompts while optimizing response times.
Using LangChain and LangGraph autonomous agents helped the conversation flow more coherently, allowing for complex decision-making paths within interactions. Modeled after a multidisciplinary team of subject matter experts, this multi-agent architecture was structured as a graph of agents, each specializing in a different aspect of the workflow. LangGraph’s directed graphs enabled flexible routing of tasks between agents based on the user’s input and the context of the conversation.
Agents were also given power to generate SQL and retrieve relevant demographics and grant data. By augmenting the LLM’s inherent knowledge with external, up-to-date data sources, the agentic bots were able to deliver precise and contextually relevant results.
By embracing the full generative AI stack, the foundation has set a new standard in grant analysis and reporting, allowing for real-time, data-rich insights and adaptive decision-making. This transformative approach to grant analysis puts this foundation in a position to expand its impact supporting nonprofits that deliver critical educational opportunities and enabling faster, more effective funding decisions.