AI Data Transformation for Funding Opportunities

Ingests raw web data

Extracts structured outputs using AI

Validates and filters results

Scores and stores high-quality opportunities

Opportunities are scattered across unreliable sources

Search results lead to generic listings, not specific calls

Data is unstructured and difficult to compare

High levels of noise (expired, duplicate, low-credibility)

The system needed to:

Aggregate funding data from multiple web sources

Transform unstructured content into structured outputs

Improve result relevance and credibility

Enable efficient exploration and evaluation of opportunities

The system was designed as a scalable, end-to-end pipeline that transforms unstructured web data into structured funding opportunities.

It integrates search, scraping, AI extraction, validation, and storage into a unified workflow.

I iteratively refined prompts to improve extraction quality, focusing on precision, completeness, and reducing noise.

This process ensured consistent, reliable structured outputs from unstructured web data.

To ensure data quality, I implemented a multi-step validation system that filters out expired, low-credibility, and irrelevant opportunities before they reach the user.

The user experience was designed around real-time interaction, allowing users to monitor progress, control generation, and explore results dynamically.

Instead of static loading, the system streams results incrementally, showing live progress and allowing users to stop generation once sufficient results are found.

Structured the UI using Aston University’s branding, translating colours, typography, and visual elements into a consistent, component-based design system. This ensured alignment between design and implementation through tokens, reusable components, and documented usage patterns.

Users can refine results through filtering by department, category, and closing date, alongside keyword-based search.

To complement structured search, a conversational interface was introduced to support natural language exploration, enabling users to discover opportunities beyond predefined filters.

Introduced a dark mode option based on user feedback, enhancing readability and supporting extended use in data-heavy workflows.

A more detailed case study, including system architecture, AI workflows, and implementation decisions, is available upon request.