IASSIST 2017 hackathon lessons for Australian food security

The IASSIST 2017 hackathon was held within “Data in the Middle: The Common Language of Research” at the University of Kansas in Lawrence from 23 to 26 May 2017. Its focus was solving food security problems by turning fragmented information into useful action: whether families can obtain nutritious food reliably, whether producers can respond to risk, and whether policymakers can identify gaps before a crisis becomes urgent. The archived program brought together work on big data, deep learning, digital agriculture and global food security, giving the hackathon a distinctly practical setting rather than treating data as an end in itself.

For Australian audiences, that focus remains highly relevant, although the operating environment differs from Kansas. Australia combines productive farming and sophisticated logistics with long supply routes, high housing costs, pressure on household budgets and substantial differences between metropolitan and remote communities. The most transferable lesson is methodological: bring data from research, government, industry and community groups into a shared framework, then design decisions around the people who need information. Applied well, predictive analytics, remote sensing and open datasets can help identify food-access needs, strengthen agricultural decisions and make scarce resources more targeted.

From shared data to food-security action

From the outset, IASSIST 2017 treated “data in the middle” as a bridge between disciplines and institutions. Plenary themes, parallel presentations and the hackathon all pointed toward common vocabulary, interoperable workflows and collaboration among people who would not normally share datasets. The archived schedule also included practical guidance on check-in, transport and local recommendations, a reminder that usable systems depend on clear processes and familiar user journeys as much as sophisticated models. In a food-security project, the central question is not whether a dataset exists, but whether a grower, welfare worker, dietitian or policy analyst can act on it.

Dimension	IASSIST 2017 setting	Australian setting	Shared design requirement
Food geography	Research and extension networks centred on Kansas	Long national supply chains and widely dispersed communities	Location-specific indicators and locally meaningful thresholds
Data partners	Universities, researchers and conference participants	Producers, state agencies, FSANZ, retailers and community organisations	Clear ownership, consent and data-sharing agreements
Decision setting	Regional research and prototyping	Metropolitan, regional and remote service delivery	Outputs designed for users with different levels of access and digital literacy
Analytical tools	Big data and deep learning presented across the program	Satellite imagery, weather records, market data and machine learning	Explainable methods, documented assumptions and quality checks
Desired outcome	Actionable research on food security	Better availability, affordability, nutrition and system resilience	Measurable decisions and outcomes, co-designed with local participants

The comparison shifts attention away from copying a conference model wholesale. A prototype that identifies a hunger hotspot in one community may need entirely different inputs in a place where households travel long distances for groceries or where seasonal conditions alter yields. Australian teams should begin with a defined decision—who will use an output, what action follows and which outcome matters—then assemble the minimum credible data. Interoperability should cover identifiers, location, dates, units and definitions of food access. Without that discipline, a polished dashboard can conceal incompatible assumptions and produce false confidence.

What the hackathon demonstrated

The strongest interpretation of the hackathon is that food security is a data-coordination problem as well as an agricultural or social problem. Food availability depends on production and distribution; access depends on price, transport and eligibility; nutrition depends on what people can actually afford and prepare; and stability depends on climate, labour, trade and infrastructure. Big-data techniques can reveal patterns across those dimensions, while deep learning can classify images, estimate yields or detect changes. The conference’s value lay in connecting these methods with substantive questions and in showing how a small team can move from a messy brief to a testable workflow.

Prototypes also need boundaries. Training data can reproduce geographic bias, correlations can be mistaken for causes, and automated recommendations can disadvantage households with limited connectivity or digital literacy. Community-controlled information about food programmes, health or school meals requires careful consent and governance. The practical information archived with IASSIST 2017 models another kind of inclusion: a participant should know where to check in, how to travel, what local services exist and how support is organised. Australian adaptations should combine technical validation with plain-language outputs, accessible interfaces and human review, ensuring that automation supports rather than replaces local expertise.

Australian realities that shape food access

Australia’s geography makes distribution a defining variable. Produce moves through long chains linking farms, processors, wholesalers and retailers, while people in places such as Perth, Adelaide, Melbourne and Sydney may have several store options. Outside metropolitan centres, including many remote Aboriginal and Torres Strait Islander communities, transport can be costly, unreliable or seasonal. A map of store distance is therefore incomplete without information about public transport, road access, delivery costs and local food production. Remote communities may rely on a mix of store-bought food, community programmes, fishing, gardening and culturally important foods, so availability must be understood locally rather than inferred solely from supermarket counts.

Shopping habits and the domestic market shape a second reality. Coles and Woolworths anchor grocery retail in many metropolitan areas, while independent grocers, farmers’ markets, food banks and community organisations are important in smaller locations. Urban households commonly combine a large weekly shop with smaller convenience purchases, increasingly through online ordering or click-and-collect, while budget stress makes price, package size and waste more influential. Food insecurity includes compromised nutrition and skipped or reduced meals, not only an absence of calories. Surveys, service records and anonymised transaction data can be combined, with privacy safeguards, to map where affordability and availability diverge.

Legal and biosecurity conditions add further constraints. Food-safety standards in Australia are developed through Food Standards Australia New Zealand under the Food Standards Act 1991 and the Australia New Zealand Food Standards Code, with enforcement shared by states and territories. Imports and agricultural goods are also governed by the Biosecurity Act 2015 and related import conditions. These rules mean a food-system application must distinguish consumption data from compliance data and must not recommend produce that lacks lawful clearance. Digital prototypes should also account for the Privacy Act 1988 where applicable and community expectations about control of information. Compliance is not a final box to tick; it determines what can lawfully be collected, shared and deployed.

Building prototypes for local decisions

An Australian adaptation should begin with a narrow, high-value decision and assemble a data-sharing compact around it. A crop-risk prototype might combine satellite imagery, rainfall, soil information, pest alerts and grower observations. A food-access prototype might compare anonymised service data with transport links, retail locations, local price surveys and community knowledge. Machine-learning models should be judged by forecast accuracy, calibration and usefulness, not simply by the number of variables. Explainable outputs, quality checks and a clear refresh schedule make predictive tools easier for public agencies, farmers and community workers to trust and challenge.

Co-design determines whether the prototype survives beyond the hackathon. People familiar with a place can identify false positives, culturally unsuitable assumptions and referral pathways that formal datasets miss. Teams should include primary producers, public-health professionals, food-distribution workers, data stewards and residents with lived experience, then compensate community participants for their time and expertise. Open formats, shared vocabularies, documented APIs and privacy-preserving analysis reduce dependence on one vendor or researcher. A credible plan also names the owner of each dataset, the cost of maintenance and the decision a weak forecast will trigger, including human escalation or a conservative default.

A practical agenda for Australian collaborators

The IASSIST 2017 example points to a practical agenda for teams working across Australian states, territories and sectors:

• Start with a food-access or production decision tied to an identified user, place, timeframe and measurable outcome.
• Establish shared definitions for food insecurity, affordability, nutritional quality, geographic coverage and forecast confidence, using consistent units and metadata.
• Pair quantitative information with local knowledge through paid co-design, particularly in regional and remote communities.
• Test models for geographic fairness, reliability and accessibility before operational use, and document limitations alongside results.
• Publish non-sensitive methods and governance documents so public agencies, researchers and producers can reuse components.
• Build a funded transition from prototype to service, including data maintenance, staff training, cybersecurity and human review.

Success should be evaluated through changed decisions rather than the number of pilots. Relevant indicators might include earlier identification of service pressure, reduced food waste, faster referral to support, more resilient production decisions and clearer communication to households. Funding bodies can require teams to report participation, data quality, model performance and operational outcomes in one release. This keeps the focus on practical impact while making inequitable effects visible. The hackathon’s enduring lesson is that shared language, disciplined data and local judgment can convert research into a stronger food system without pretending that a model can resolve political, economic or environmental uncertainty on its own.