Talk to us
New
Affinda Gateway
Affinda Extract
About Affinda
Leadership
Careers
Media & Awards
Draftable
Certification & compliance
Trust centre

For students

We build AI products, develop custom systems, and help businesses put AI to work.

See what we’re building

Across Affinda

Document automation

Software that reads documents and extracts the data in them.

Explore Affinda Extract
  • Document processing
  • Data extraction
  • Resume parsing
  • Data validation
  • Straight-through processing

Deployment – Applied AI engineering

Custom AI systems built for individual organisations.

See the work
  • AI agents
  • Workflow automation
  • Knowledge assistants
  • Voice AI
  • Demand forecasting

Deployment – Strategy and value creation

AI training and adoption programs for leaders and teams.

See the work
  • AI education
  • AI adoption
  • Opportunity discovery
  • Pilot development
  • Building AI capability

From our former interns.

Jamie

AI Analyst
Doing a Politics, Philosophy, and Economics degree, I assumed tech roles would be out of the question for me. Affinda proved me wrong. I was given the opportunity to build an internal performance review app that the company has now used to run its performance reviews. It is a rare and special opportunity to be given that much trust and agency over a real project.

Bachelor of Politics, Philosophy and Economics

Monash University

Winter intern, 2026

Alexis

Product Specialist
At a time when AI is rapidly transforming how we work, my internship at Affinda gave me a rare opportunity to explore creative new approaches to problem-solving. The experience shaped my outlook on AI: if you can think it, you can do it. Working alongside the team at Affinda fostered this ambitious mindset and taught me lessons I will carry with me for life.

Juris Doctor

University of Melbourne

Summer intern, 2025

Jack

AI Engineer
A new AI capability can launch at 6 am, and by that morning we’re already testing what it makes possible. I get to explore those questions with some of the smartest people I’ve met, with a lot of trust and freedom to pursue my ideas. You’re encouraged to be inventive and ambitious in what you take on. I find that incredibly exciting. If that sounds like your kind of thing, come and see for yourself.

Master of Data Science

University of Melbourne

Summer intern, 2025

Liam

AI Solutions Analyst
In just six weeks at Affinda, I researched agentic AI and built working automations and agentic workflows — an area I didn't know existed before joining. Coming from a non-technical background, the learning curve was steep, but support from a diverse team of passionate and ambitious people made it very rewarding. I gained hands-on experience applying frontier AI tools to real business workflows.

Bachelor of Commerce

University of Melbourne

Summer intern, 2025

A few questions we’ve explored.

Interns explored each of these with guidance from our team and room to try their own ideas.

Scroll to explore

Agent systems

Can AI models make each other better?

Different models bring different strengths. How should they divide the work and build on each other’s ideas?

AI agents can write code, research a market or work within a product. Combining their different strengths can help them achieve more together than they could individually.

We’re interested in how to make that collaboration work: how agents share context, divide responsibilities and build on each other’s ideas. The way they’re organised can matter as much as the capabilities of each agent.

Multi-agent systemsPrompt designEvaluationPython
A different perspective changes the idea. A different perspective changes the idea. Different strengths. Something built together. GPT Astra Fable 5.1 What if… Or try this? Explorer Builder Challenger Coordinator Try another way
A different perspective changes the idea.

Machine learning

Which model should handle this task?

The challenge is knowing when a smaller model is enough.

Different tasks call for different levels of capability. We’ve explored how to delegate work intelligently between models, using a smaller, faster model when it can do the job and more reasoning power when the task needs it.

The tricky part is recognising difficulty before asking a model to answer. Prompt length tells us very little. A model’s internal representation of a question may offer a more useful clue.

Model evaluationRepresentation analysisExperiment designPython

How can AI tell how hard a question is?

Choosing a model means matching its capabilities to the task. A smaller model may be enough, but we need a sense of difficulty before it starts answering.

Prompt length won’t tell us. “2 + 2?” and “P = NP?” are both short, but ask very different things. Could a language model recognise that difference internally?

As a model processes text, its layers build representations of what it is being asked. We looked at the mid-layer activations of Qwen3-4B: long lists of numbers that describe its internal state partway through processing a prompt.

Using PCA, we projected those representations into two dimensions. In this sample, easier and harder prompts tended to fall on different sides of a rough boundary, with some overlap. That pattern suggests a possible way to estimate difficulty before deciding which model should answer.

Swipe across the graph to explore it.

A signal of difficulty inside Qwen3-4B A projection of internal model activations for 20 prompts. Easier and harder questions tend to fall on different sides of a rough boundary, with overlap. A signal of difficulty inside Qwen3-4B easierharder -40-30-20-10010203040 -40-30-20-100102030 ← easierMid-layer activations · PC1harder → PC2 Say hello. Capital of France? What is 2 + 2? P = NP? Riemann Hypothesis Prove Fermat's Last Theorem easymediumhard
Not every task needs the biggest model. Not every task needs the biggest model. Use a smaller model when it can do the job. More reasoning can help. It also costs more. What’s 2 + 2? Debug thisrace condition Router SmallFocused tasks $0.001illustrative cost / task GeneralBroad capability ReasoningComplex tasks $0.10illustrative cost / task
Not every task needs the biggest model.

AI for sales & marketing

Can AI spot the right moment to reach out?

Understand a potential customer, what they need, and when our products could help.

Understanding a potential customer means bringing together what a company does, what its people are talking about and what is changing in the business. AI agents can research those signals and connect them into a useful picture.

That context can help us identify who might benefit from our products, why they might find them useful and when to start a conversation. It’s one example of how we use AI across the business, including marketing and sales.

AI agentsCompany researchAutomationSales and marketing
Research agents investigate a potential customer. Research agents investigate a potential customer. A synthesis agent connects their findings. Who to contact. Why them. Why now. Potential customer People agent LinkedIn Operations lead flags admin overload Hiring agent Open roles Growing the operations team Company agent Company news Expanding into new markets Agent recommendation Contact the operations lead They raised the admin problem Their team is hiring Expansion means more work Start with what could help: AI that takes admin off their hands. Sales team Synthesis agent
Research agents investigate a potential customer.

Bring your energy
and ideas.
  • Understanding a business
  • Making complex ideas clear
  • Testing an idea
  • Talking to customers
  • Researching a market
  • Teaching people how to use AI
  • Building a product
  • Finding out what works

Ways to get involved

  • Summer internship

    Build something useful, with training and support from our team. Take ownership of your project and learn alongside other interns. Applications are not open yet.

    Our team brings experience from fields such as technology, consulting, capital markets, law, commerce, scientific research and philosophy.

  • Hackathon

    An Affinda hackathon is coming. Date and details to follow.

Stay in the loop

Leave your email. We’ll tell you when internship applications open, when the hackathon has a date, and about other student events.

What’s coming

  • Summer internshipApplications are not open yet.
  • HackathonDate and details to follow.