Inspiration
With the rise of ESG reporting, the Australian government has started mandating climate compliance reports through governmental criteria (the AASB S2). This mandate has begun phasing in for large companies, continuing to roll out for smaller companies by 2028.
While big companies have the resources to meet compliance requirements, medium to smaller companies need hundreds of thousands of dollars to prepare climate disclosures. This is because most businesses don't have in-house ESG expertise and end up paying consultants to bridge the gap.
We looked at the market and found a polarising divide: there's expensive enterprise software built for companies that already have sustainability teams, or outsourced consultants doing the work by hand. There was nothing in between for a company that just wants a fast and first read on where they stand with the AASB S2 criteria.
What it does
GreenScreened reads a company's existing internal documents and checks them against Australia's AASB S2 standard. The app breaks the report into sections, classifies each section against the relevant pillar of the TCDR framework using an LLM, and returns a structured result with a confidence score and the exact source page every claim came from
How we built it
- Extraction & classification pipeline (Node.js): parses the PDF page-by-page, chunks the text, and sends each chunk to Google's Gemini API with a carefully engineered prompt
- Rate limiting & retry logic to handle free-tier API limits gracefully during testing.
- SQLite database to store classified results per company per year
- We sourced actual 2025 climate disclosure reports from six real ASX-listed companies to make sure our classifier generalises across industries
Challenges we ran into
The main challenges that we ran into looked like:
- Free-tier API rate limits meant we had to build proactive request spacing and retry-with-backoff, rather than firing calls as fast as possible
- Reading long documents with the API led to long wait times for testing, and if the document was too long, we risked important information getting lost.
- We were tempted to focus on all the pillars of ESG: environment, social and governance
Accomplishments that we're proud of
- Building an AI pipeline that works and can validate real data in the way we want
- We were able to do this in under 48 hours and deploy successfully
- We identified a problem that is specific and not too broad, allowing us to make something that addresses the problem head on
What we learned
Coming from different coding backgrounds, we learnt how to create an AI pipeline and developed further knowledge on all our platforms.
What's next for GreenScreened.
- A Red/Amber/Green scoring layer that turns raw pillar classifications into a completeness assessment per company.
- Deploy to different companies and test across even more industries
Log in or sign up for Devpost to join the conversation.