Care staff used to spend the first half-hour of every outreach session crowding around a printout, juggling befriending visits, R-tier welfare checks, and SGO prospects on three separate lists. CaritaHub Outreach reads all three at once, scores them by risk and recency, clusters them into a walkable route, and hands the staff a phone-shaped session that captures every outcome on the go.
Befriending pairs come from the volunteer roster. R1/R2 follow-ups come from the RIM tier review. SGO prospects come from a referral spreadsheet. Each lives in its own tab.
The senior staff "just knows" Blk 537 is next to Blk 539. New officers don't. Walking back and forth across the estate burns 20–30 minutes every session.
Notes go in a notebook, get typed up at 7pm, then sit unread until the next case-management review. Referrals slip. Conversions don't get logged.
CaritaHub treats every outreach session as a mix of three job types. The lane is the load-bearing signal — it sets the pin colour, the urgency, the question script, and the flow when the staff arrives.
Trigger: RIM score R1 or R2 with no visit in the last 7 days.
Welfare check. Front-loaded in the priority weights. The session card shows the red badge, the "RIM R3 · welfare priority" header, and any care flags (falls risk, lives alone, caregiver stress) in coral.
Trigger: Volunteer–senior pair, scheduled visit due.
Relationship maintenance. The senior knows their volunteer; continuity matters more than urgency. The session card shows the assigned befriender's name and how many visits they've had so far.
Trigger: Non-member on the outreach list, not yet contacted.
First-contact conversion. The goal is to move from prospect → contacted → converted member. Outcome panel adds a "Convert to member?" toggle for warm leads.
Lane identity is consistent across every screen: pin colour on the map, badge on the stop card, border on the session header, segment colour on the session-impact donut.
Tap through the workflow on the right. Or open the staff app live and try the full session yourself.
A four-step pipeline runs locally on the centre's AIBox. No senior data leaves the building. Mock data is from real Bedok addresses.
Pulls members with R1/R2 status untouched in 7+ days, befriending pairs due for a visit, and SGO prospects within 1.5km of the centre.
(rim × rim_weight) + (recency × recency_weight) + (lane_weight from staff focus) — staff control the lane weights via the AI focus preset.
Resolution 10 (~65m hex) groups same-block and adjacent-block candidates so that one walking trip yields multiple visits.
Nearest-neighbour with origin pinned at the AAC. Outputs an ordered stop list and arrival times that fit the staff's chosen window.
The pipeline runs on a Mac Studio M4 Max in the centre's network closet. Voice-note transcription uses on-device Gemma. Senior data never leaves Singapore.
Switch between staff, watch progress in real time, and read the H3 weekly heatmap to spot under-served blocks.
Click any name in the sessions list — the map re-centres on their zone, draws their walking route, and shows their current stop with a pulse.
Hex-binned visit counts across Bedok / Tampines / Pasir Ris. Filter by lane and date range to see which blocks need more attention.
4-week stacked trend, outcome donut, lane mix per staff, top-visited blocks. Everything reads against the same session log the field staff produce.
Every senior across every staff route. Filter by lane / RIM / staff. Last-visit recency rolls up automatically from completed sessions.
Three live surfaces. Open them on a laptop, project to the room, and walk your team through a full session.
The 4-screen field flow. Best viewed in a phone-shaped window or on an Android device.
Open staff app →Live map, sessions list, weekly heatmap, reports. Best at 1440px+.
Open dashboard →Send the link to your colleagues and walk through it together.
Back to top ↑