Granted – An AI-powered advisory platform for assessing public funding opportunities
Granted set out to simplify access to public funding for companies operating in the agriculture and food sector, where information about grants and subsidies is often fragmented, complex and difficult to interpret. To explore the technical feasibility of an AI-driven advisory platform, Granted partnered with DO OK to design and deliver a proof of concept focused on automated data collection, document analysis and eligibility assessment. The project examined how publicly available grant documentation could be processed at scale and matched against user profiles, laying the technical groundwork for a potential future MVP within a clearly defined time and budget scope.
About Our Client
Granted is a Polish technology company building an intelligent advisory platform focused on public funding for the agriculture and food sector. The company supports organisations involved in agricultural production, food processing, trade and related services by helping them navigate a complex and highly distributed funding landscape.
Although public financing opportunities are widely available, information about grant programmes is published across many official sources and often presented in formats that require specialist knowledge to interpret correctly. For many organisations, understanding eligibility rules, evaluation criteria and competitive positioning at an early stage is both time-consuming and uncertain.
Granted addresses this challenge by aggregating grant information from official public sources and presenting it in a structured and accessible way. Users provide information about their organisation and planned investment, after which the platform filters available programmes and supports eligibility verification. The system enables early-stage self-assessment, helping organisations focus their attention on opportunities that align with their profile and objectives.
The platform combines domain expertise in public funding with AI-driven document analysis and natural language processing. Its goal is to support better decision-making before organisations invest significant time and resources into the application process.
To learn more about Granted, click here.
Client Needs
During the early development of the Granted platform, the team was exploring how an AI-based advisor could support organisations in selecting suitable grants and subsidies. The objective was to analyse publicly available documentation and relate it to users’ business profiles, producing structured insights to support informed funding decisions.
As part of this exploration, the client investigated several technical approaches internally. These early efforts helped clarify the scale and complexity of the problem, particularly when working with large volumes of unstructured public data published in varying formats and updated at different frequencies. Some areas, such as automated interpretation of eligibility criteria and scoring rules, required deeper expertise in AI, NLP and scalable data processing to reach the desired level of consistency.
At this stage, the client was looking to validate alternative approaches and define a more robust technical direction for the data and AI pipeline. This included consulting on web scraping strategies, document processing methods and the selection of suitable technologies and frameworks. Scalability was an important consideration, as the system needed to support thousands of grant programmes and associated documents, many of which change regularly.
In parallel, the client wanted to validate feasibility within a limited and clearly scoped engagement. The agreed objective was to deliver a working proof of concept within a short timeframe, using real grant datasets to generate early performance indicators that could inform planning for a potential future MVP.
Why DO OK
Granted selected DO OK based on recommendations from previous clients as well as our expertise in AI- and data-intensive projects. The client was looking for a partner who could review existing assumptions objectively and propose alternative approaches grounded in practical delivery experience.
During the initial consulting phase, we worked closely with the Granted team to analyse existing concepts and explore multiple solution paths. Rather than promoting a single predefined architecture, we outlined different technical approaches and discussed their implications in terms of accuracy, cost and scalability. This collaborative process helped clarify trade-offs and supported decisions aligned with the project’s scope and constraints.
Budget considerations played an important role, as funding was secured only for the first phase. We therefore focused on solutions that were cost-efficient at the proof-of-concept stage while remaining extensible if the project were to progress further. This included recommendations around cloud infrastructure options and the use of locally deployed language models where appropriate.
Another important factor was our experience in delivering proofs of concept within short timeframes. The ability to plan, execute and validate a technically meaningful PoC within two weeks aligned well with the client’s goals for the initial phase and supported a focused, feasibility-driven engagement.
Languages
Python
Frameworks & Libraries
FastAPI
Celery
LangChain
Crawl4AI
Infrastructure & Storage
AWS S3
Docker
AI and NLP
Ollama
GPT
Project Overview
The project focused on processing publicly available grant and subsidy documentation in order to match user organisations with relevant funding opportunities. The objective was to analyse large and diverse datasets and compare them against user-provided information in a structured and repeatable way, supporting early-stage assessment before organisations invest time in preparing applications.
The first phase involved collecting data from a wide range of official public sources. Approximately 10.000 grant programmes were identified, each containing descriptive metadata such as business scope, application periods and eligibility rules. In addition to this structured information, many programmes included detailed documentation published as PDF or Word files. These documents contained extensive eligibility criteria, evaluation rules and formal requirements that needed to be extracted and interpreted as part of the assessment process.
Once collected, the data was normalised and transformed into curated assessment criteria. This step aimed to translate complex, often legalistic programme documentation into a form that could be evaluated consistently across different grants.
Evaluation was performed in two stages. First, hard criteria were applied to determine whether participation in a given call was feasible at all. Second, soft criteria were used to estimate potential scoring outcomes in competitive programmes, reflecting how well a given organisation’s profile aligned with the programme’s evaluation framework.
The output of this process was a structured report presenting a ranked list of grants relevant to the user. Each entry included eligibility context and assessment results derived directly from the processed documentation, together with supporting information to help users understand why a given programme was recommended. This approach enabled early, informed decision-making and provided a clear foundation for further development beyond the proof-of-concept stage.
Challenges and Solutions
One of the main challenges was handling large volumes of unstructured data originating from multiple public sources, each with different update frequencies and document formats. To address this, we implemented parallel scraping and processing using a distributed task queue, allowing data to be collected and processed efficiently while supporting frequent updates.
Balancing processing quality with cost efficiency was another important consideration at the proof-of-concept stage. The solution combined classical natural language processing techniques with large language models, while design decisions were made to keep operational costs appropriate for an exploratory phase. Where possible, we proposed the use of locally deployed language models instead of commercial APIs.
Document processing introduced further complexity due to variation in file formats and content quality. Several approaches were explored, including text extraction, Markdown transformation and OCR. For OCR, Tesseract was selected as a proven and cost-effective solution suitable for the PoC scope, with more advanced approaches identified as potential options for later phases.
Given the exploratory nature of the project, data availability and network reliability were handled pragmatically through basic retry mechanisms and fallback strategies. This ensured continuity of processing without introducing unnecessary architectural complexity at an early stage.
Impact and Outcomes
The project concluded with the delivery of a functional proof of concept within a 2-week timeframe. The PoC covered approximately 20–30% of the identified grant datasets and enabled daily ingestion of updated grant data into the processing pipeline. It was deployed to a test environment, allowing the client to validate the solution using real-world data and representative scenarios across different types of grant programmes.
At the proof-of-concept stage, accuracy was measured at around 60–70%. Given the complexity, variability and unstructured nature of publicly available grant documentation, this result demonstrated that the chosen technical approach was effective and well suited as a foundation for further refinement. The outcomes confirmed that large volumes of heterogeneous grant data could be processed, normalised and assessed in a consistent way using the proposed pipeline.
Beyond validating the core technical assumptions, the PoC provided valuable insight into performance characteristics, data quality challenges and areas for further optimisation. The engagement was intentionally focused on feasibility validation and successfully established a solid technical foundation for future development, supporting informed decisions about potential next phases of the product and the path towards an MVP.
Working with DO OK gave us the clarity we needed to assess the feasibility of our AI-driven platform and make informed decisions about the next stage of product development.
Founder at Granted Fund
