← All projects
Product strategy · Business analysis · AI workflow design

Rumi

A human-supervised decision system that turns structured rental enquiries into room recommendations for Leasing review.

Role
Product & BA / Delivery lead
Team
Nicole + 1 developer
Status
Shipped · 2026
Timeline
3 weeks to build

I led the Leasing discovery, designed the decision logic and human hand-off, and worked with one developer to ship the workflow.

The next problem appeared downstream

The Guided Enquiry flow on the RoomingKos website improved the quality and structure of incoming rental enquiries. But room matching and recommendation were still manual, typically taking Leasing 5–7 minutes per enquiry.

Guided Enquiry
Better, more structured prospect information.
→
New bottleneck
Room matching and recommendation were still manual.

What Leasing’s process revealed

Room matching relied on a mix of fixed business rules and contextual judgement. Some criteria were clear-cut, while preferences such as location, budget and room type could carry different weight depending on the prospect’s wider needs and trade-offs.
Some requirements could not be verified from the available data. Those unknowns needed to remain visible rather than being guessed away.
The final recommendation still needed a human owner. A system could narrow and explain the strongest options, but Leasing needed to retain control of the decision and customer response.
How do we structure the way an experienced Leasing Manager evaluates and recommends rooms — without removing the judgement that still belongs to them?
The question behind Rumi

See the decision logic in action

A simplified recreation of how Rumi filters available rooms, compares viable options and prepares a recommendation for Leasing review.

RK
RoomingKos
Rumi for Sales
Today Enquiries Properties Admin
Recreated demo · not live
1
Enquiry
Applicant
Student
Budget
$350–450 / wk
Preferred area
Monash Clayton
Move-in
10 Feb
Room type
Single, solo
Mandatory
Female-only
2
Matches
FEASIBILITY CHECKS — Lease · Category · Occupancy · Move-in window · Location eligibility.
REQUIREMENT CHECKS — must-have requirements are enforced where the data allows, or surfaced for human verification where it doesn’t (see Riverside, below).
{{ pick.rank }}
{{ pick.name }}
{{ pick.priceLabel }} / wk
{{ pick.why }}
Why matched
3
Reply
{{ replyCount }}
SEND AS
FORMAT
NOTES
Built from a template I wrote — Rumi fills in the room facts and adapts tone or emphasis to staff instructions.
{{ appliedTag }}
{{ replyText }}

How I designed the decision system

The design decisions behind the workflow.

01 Separate rules from judgement
{{ label2 }}
I modelled the recommendation process in layers: remove impossible options, compare viable ones, surface uncertainty, then use AI where contextual reasoning adds value.
Prospect needs Student · Oct move-in · 12+ months · $400/wk · Monash Clayton · Female-only
{{ stageQuestion }}
{{ row.a }} {{ row.b }}
{{ stageResult }}
{{ stageNote }}
Impossible → Remove Trade-off → Compare Unknown → Verify
02 Keep the human in control
{{ label3 }}
I designed the hand-off so Leasing could inspect the evidence, change which rooms were included and tailor the final response before anything reached the prospect.
Rumi prepares
Summary · Recommendations · Rationale · Trade-offs · Unknowns
↓
Leasing reviews
Verify · Include / Exclude · Adjust recommendation
↓
Rumi drafts response
Email / WhatsApp · Standard / Concise / Bullets
↓
Leasing customises
Edit directly · Add instructions · Regenerate
↓
Final customer response
Reduce cognitive work — without transferring decision or communication control.
03 Design for uncertainty and failure
{{ label4 }}
The workflow needed explicit behaviour when normal decision-making broke down. I designed paths for model failure, missing data and overlapping ownership so Rumi could fall back, flag or hand over instead of silently failing or inventing certainty.
{{ node.text }}
04 Design beyond the feature
{{ label1 }}
I treated Rumi as one layer in a wider customer and operational system rather than a standalone AI feature. Structuring the enquiry first, then the decision that followed, created a reusable foundation that could later connect into CRM and operational workflows.
Enquiry → Recommendation → Conversion → Operations
Future direction — CRM and operational connections are not yet built.

The system doesn’t have to know everything. It should be clear about what it can decide, what it should surface, and when human judgement should take over.

DECIDE → SURFACE → HAND OVER

Outcome

Room matching & response preparation
Rumi cut room-matching and response preparation from 5–7 minutes to under one minute, generating both a review-ready recommendation and a prospect-ready response for Leasing.
5–7 min
<1 min
manual analysis → review-ready output
01
Repeatable decision flow
Room matching became a repeatable decision flow, with rules, trade-offs and unknowns surfaced consistently for review.
02
Faster follow-up
Leasing receives both the recommendation and a prospect-ready response, reducing the work between enquiry and customer follow-up..
03
Designed to extend
The structured enquiry-to-decision flow provides a reusable foundation for future CRM and operational integrations.
AI workflow Decision rules Human-in-the-loop Failure design
← Back to all projects