Yarning Connect

Inspiration

Yarning was born from a simple observation: across Australia, Indigenous communities face critical gaps in health, housing, education, and employment — yet companies with RAP commitments and ESG budgets struggle to find where their support is actually needed. Skilled professionals want to help but have no visibility into which communities need their expertise. The disconnect isn't a lack of resources — it's the lack of a system that connects them effectively.

We were inspired by the Indigenous tradition of "yarning" — sitting in a circle, sharing knowledge, and building understanding. We wanted to build a digital platform that honours that same principle: bringing communities, companies, and professionals into conversation so that support reaches the right place at the right time.

What it does

Yarning is a real-time, season-aware platform that connects Indigenous Australian communities with corporate and professional support through four AI-powered engines:

Urgency Scoring Engine — Communities submit proposals describing their needs. After on-the-ground verification by the Yarning team, our AI converts these into structured urgency scores across six weighted categories (Health, Food & Water, Housing, Education, Employment, Cultural Programs), adjusted by a Neglect Multiplier and a Seasonal/Timeliness Modifier. This produces a single 0–100 urgency score that updates dynamically.
Auto-Matching Engine — Our AI evaluates every company and professional against all community profiles, calculating a compatibility percentage based on needs-focus alignment, funding capacity, RAP commitment, contextual fit, and strategic synergies. Companies see their best-fit communities; communities see which supporters match their priorities.
Dynamic Readjustment — Urgency scores are living numbers. Every time support is delivered, the AI recalculates the affected category scores downward, ensuring resources flow to wherever the need is greatest right now, not based on outdated assessments.
Impact Report Generator — After support is delivered, the AI generates comprehensive before-and-after impact reports covering score changes, measurable outcomes, community voice, and RAP target alignment — ready for board presentations and public reporting.

All of this is visualised on a colour-coded urgency map of Australia, giving instant national visibility into where support is needed most.

How we built it

We built Yarning as a React (Vite) + Node.js/Express application using Tailwind CSS for a clean, professional UI. The AI features are powered by the Gemini 2.0 Flash API, which handles natural language analysis of community proposals, urgency scoring, match calculation, score readjustment reasoning, and impact report generation.

Each AI feature sends structured prompts to Gemini with the relevant community and company data, and receives JSON responses that are processed by our backend and rendered in the frontend. The scoring model uses a weighted formula across six need categories with neglect and seasonal multipliers, while the matching engine evaluates five weighted compatibility criteria.

We designed the platform with cultural sensitivity at the centre — incorporating Indigenous seasonal calendars (e.g. Dharrathamirri, Midawarr) into our seasonal intelligence rather than using Western calendar seasons, and framing communities by their strengths and aspirations rather than deficit language.

The prototype uses fictional but realistic data representing four communities across NT, SA, and WA, and five major Australian companies (Wesfarmers, BHP, Commonwealth Bank, Rio Tinto, Qantas) with real RAP tier structures.

Challenges we ran into

Designing a respectful scoring system — Reducing complex community needs to a number is inherently reductive. We spent significant time ensuring the model accounts for context through the neglect multiplier and seasonal modifiers, so scores reflect reality rather than oversimplifying it.

Prompt engineering for consistency — Getting Gemini to return reliably structured JSON with realistic, non-inflated scores required extensive prompt iteration. Early versions would rate every match at 85%+ or give every category a 5/5 — we had to carefully instruct the model to be conservative and realistic.

Cultural sensitivity in AI outputs — Ensuring the AI avoids deficit framing when describing Indigenous communities was a deliberate design choice that required careful system prompting. The AI needed to centre community strengths and agency, not just list problems.

Seasonal intelligence — Integrating Indigenous seasonal calendars meaningfully (not just as labels) required research into how seasons like Dharrathamirri actually affect community needs — water availability, ceremony timing, construction feasibility — and translating that into scoring adjustments.

What we learned

We learned that AI can be a powerful tool for social impact when it's designed with the right constraints. Unconstrained, the model produces generic, overly optimistic outputs. With careful weighting, multipliers, and culturally informed prompting, it becomes genuinely useful for prioritisation and resource allocation.

We also learned that the hardest part of building social impact technology isn't the code — it's the design decisions around how to represent real human needs in a system without reducing people to data points. The seasonal intelligence and community voice features exist specifically to keep humanity in the platform.

What's next for Yarning

Real community partnerships — Moving from fictional data to co-designing the platform with actual Indigenous communities and organisations, ensuring the tool serves their needs and governance structures
Professional matching — Expanding the matching engine to connect individual skilled professionals (doctors, teachers, engineers) to communities based on specific skill gaps
Longitudinal impact tracking — Building trend analysis so communities and companies can see how urgency scores change over months and years
Mobile-first community interface — A simplified mobile app for community leaders to submit and update proposals from remote areas with limited connectivity
Government and philanthropic integration — Connecting Yarning data with existing government funding databases to improve the neglect multiplier accuracy

Link Glossarium

Link 1: GitHub repository containing the source code for the Yarning platform
Link 2: GitHub repository containing the four AI engines powering Yarning
Link 3: Live Yarning platform
Link 4: Yarning whitepaper detailing the platform's features and functionality
Link 5: Research paper outlining the problem landscape, proposed solution, and supporting evidence