Humnloop / Full stack AI design studio

Human in the Loop

We build AI tools with the people who will use them.

Humnloop is a Melbourne studio. We start with a working prototype on your data and build it out once your team has used it.

Model driftHeld by review
Move across the ring to review it. Click to add a checkpoint.

01Where we start

Most AI pilots look good in a demo and then go no further.

The data usually sits in systems that were never built to connect. Privacy and security questions arrive after the tool is built. We deal with both while the first version is made, and your team uses it from the first week.

02Ways to work together

Start with a prototype, then decide what comes next.

A first project is small and fixed in scope. You end up with software your team has used on real work. If it works, we take it into production. If not, the cost was small.

  • A working version on your own data, built in a short fixed scope. Your team uses it on real work and we decide together what to keep.

    Fixed scopeYour dataWorking software
  • We take a prototype that has proved itself through your security and privacy review and into daily use, with access controls, an audit trail and documentation.

    Security reviewAudit trailDocumentation
  • An AI workspace that answers from your own knowledge and shows its sources. You decide who sees what, and each request to a model is checked against where your data may go.

    Cited answersAccess rulesData residency
  • We stay on to extend the system and add new tools on the same data.

    Next toolsSupportSame foundations

03Sprint to enterprise

Rapid prototyping to enterprise scale.

01A prototype sprint
02A product in daily use
03An enterprise system

04Selected work

Some of the tools we have built, and where they are used today.

Each card shows the problem, what we built and the result, with screens from the product.

Case studies
Swinny.ai
Swinny.ai: Marketing office workspaceSwinny.ai: Tenders and grants workspace
Marketing office workspace

01University commercial teams

One record of the people, partners and opportunities a team works with

Problem
Finding the right expert for a partner took months across disconnected systems.
Built
A business development platform with one record of people, partners and opportunities.
Impact
Teams work from one record and handle more with the same people.
Signals
Signals: Let people build with AISignals: A secure workspace for your knowledgeSignals: Every question passes your checksSignals: Your database is the boundarySignals: Run it where your data has to stay
Let people build with AI

02Universities, government and business

A self-governed AI sandbox for your own knowledge

Problem
Staff were using AI tools with no control over where knowledge went.
Built
Workspaces where people build their own agents on the organisation's knowledge, with every answer sourced.
Impact
A council association and a manufacturing research centre now use AI inside rules they set.
Segendor
Segendor: Scan a file before it reaches productionSegendor: Where a dataset is missing values
Scan a file before it reaches production

03Data, migration and privacy teams

Find what is wrong with your data before it reaches production

Problem
Data problems surfaced only after migration or AI use, and enterprise tools were out of reach.
Built
A browser scan that finds quality and privacy issues down to the record.
Impact
Teams fix their data before it reaches production.
Website analysis
Website analysis: How every page links to the rest
How every page links to the rest

04University web and marketing teams

A map of how every page on a large website connects

Problem
Nobody could see how a large university website linked together.
Built
A weekly crawl and dashboard that maps every page and broken path.
Impact
Web teams plan fixes and new structure from evidence.
City Pulse
City Pulse: Find your council and its peersCity Pulse: See where people live, street by streetCity Pulse: Look past the council averageCity Pulse: Ask the base a question in plain wordsCity Pulse: A growing library of public and council data
Find your council and its peers

05Councils

One common base for every council in the state

Problem
Council leaders gathered the same indicators by hand each month.
Built
One common base for every Victorian council, with a map and plain-language questions.
Impact
Running as a prototype, with councils invited into a pilot.
Ecosystem map
Ecosystem map: Organisations by stateEcosystem map: Supplier mapEcosystem map: Who can make what, from whatEcosystem map: Where the supply chain is thin
Organisations by state

06Industry research centre

A whole industry's capability on one map

Problem
No single view showed who researches, supplies and makes things in the sector.
Built
An interactive map of the sector's organisations, processes and supply chains.
Impact
Published by the research centre as a shared view of national capability.
Tree map
Tree map: Canopy across the council areaTree map: Summer surface heatTree map: Where more trees would cool mostTree map: A neighbourhood in detailTree map: Heat at street level
Canopy across the council area

07Councils

Every tree, canopy and hot street in a council area, in one file

Problem
Canopy, heat and tree records sat in separate data.
Built
One map that runs from whole suburbs down to a single tree.
Impact
It shows where planting would cool streets most.
ClimateArray
ClimateArray: Portfolio climate riskClimateArray: Exposure at building levelClimateArray: Loss across the portfolioClimateArray: Site by site review
Portfolio climate risk

08Banking and property

Fire, cyclone and flood risk for every asset in a portfolio

Problem
Lenders could not see physical climate hazard across every asset they hold.
Built
Fire, cyclone and flood models with a portfolio platform built around them.
Impact
Portfolio losses and disclosure come from the same data, asset by asset.

05How review works

Someone on your side reviews each stage before the next one begins.

  1. 01 / Frame

    Frame the decision

    We talk to the people who will use it and agree what good looks like.

  2. 02 / Prototype

    Build it early

    A working version on your data while the big questions are still open.

  3. 03 / Prove

    Test with real users

    Your team uses it on real work and tells us what to change. You decide what goes into production.

  4. 04 / Scale

    Harden and hand over

    What you keep is made production ready and handed to your team.

A person reviews the work between every stage

FrameStage 01 of 04

06Data and handover

What happens to your data, and what you keep at the end.

01ACCOUNTABILITY

Someone on your side signs off each stage. You can see what every answer was based on.

02PROVENANCE

Every number shows where it came from and when.

03SECURITY

Your data stays in the region you choose and is kept separate from other clients. Keys never reach the browser.

04HANDOVER

You get full documentation, and your team can run the system without us.

07Who builds it

Humnloop is led by Joseph Glesta, who has worked across government, research and startups.

Joseph Glesta took Digital Twin Victoria to market for the Victorian Government, then moved research toward commercial use at CSIRO's Data61. He co-founded ClimaSens, a climate risk company that won the Google.org Impact Challenge on Climate Innovation, and now builds AI products at Swinburne University of Technology.

He also builds rapid prototypes that multi-million dollar companies and organisations use today.

08Get in touch

Tell us what you are working on, and we will tell you if we can help.

A few lines is enough. We will reply with what a first prototype could cover and a rough cost.

What are you after