Project Description
CommunityMate
CommunityMate is an agentic AI “virtual coworker” that helps residents find government/community services, compare nearby options, get directions, and complete tasks like bookings, form submissions, and reminders safely and inclusively.
It supports low digital literacy users with plain language, voice input, and human handover.
Problem
Solution
- Agentic assistant focused only on AU government/community services (no advice; safe referrals)
- Understands free text/voice, asks 1–3 simple follow-ups (where, transport, when)
- Lists 2–5 nearby options with hours, contacts, and estimated travel (car/walk/transit)
- With consent: simulate bookings, form submissions, and set calendar reminders (.ics)
- Human handover: create a volunteer follow-up note (simulated)
Technical Architecture
- Frontend: Streamlit app with microphone capture (
streamlit-mic-recorder)
- Agent: LangChain + OpenAI Chat model (
gpt-4o-mini default)
- Semantic search: FAISS index + OpenAI embeddings (
text-embedding-3-small)
- Geospatial:
- Geocoding: Nominatim (OpenStreetMap)
- Routing: OSRM for car/walk; heuristic transit
- Tools: find services, record consent, book, fill forms, set reminders, update profile, escalate
- Data: CSV datasets from gov portals via
SERVICES_CSV_PATH
- Files: Local JSON/ICS receipts written to
./out
- Config via
.env: OSRM_BASE_URL, SERVICES_CSV_PATH, ALLOW_NON_GOV, LLM_MODEL, DEMO_MODE
Demo Screenshot
Innovation
- Agentic autonomy with trust guardrails: multi-step tasks only after consent; minimal storage with transparent receipts
- Transport-aware prioritisation: results re-ranked by walk/transit/car preferences
- Inclusion by default: voice capture, easy language, multi-language replies, and “near me” via saved home address
- Social handover: seamless “escalate to human” option supports cohesion and wraparound help
Accessibility & Inclusion
- Low-literacy support: short sentences, step-by-step options, “easy-read” tone
- Language support: reply in the user’s language; internal prompts enforce scope
- Disability-aware: voice input in demo; large-text & simplified content; roadmap includes WCAG checks
- Transport & mobility: shows walking, driving, and transit estimates; promotes nearby options
Ethics, Privacy & Trust
- Consent first: bookings/forms/reminders only after explicit consent
- Minimise data: stores only what’s needed, with local JSON receipts for transparency
- Safety boundaries: no clinical/legal advice; crisis routing (000, Lifeline, 1800RESPECT)
- Compliance: designed to align with APP & Security by Design; demo = no third-party sharing; roadmap includes DPIA/PIA and threat modelling
Scalability & Adaptation
- Works with any dataset (name/address/lat/lon) via CSV auto-detection
- Configurable across LGAs, councils, and states
- Modular tools enable future integration with booking APIs, SMS/email, translation, and human helpdesk
Roadmap
- Connect to live booking APIs for selected services
- Add secure SMS/email channels for reminders
- Service freshness checks & quality signals
- WCAG audit & translations for priority languages
- On-device/edge inference for sensitive use cases
Impact (Metrics)
- Task completion: % users finding a service + next step (map open, reminder, handover)
- Time-to-solution: median ask → action time
- Inclusion: uptake by low-digital-literacy users (via easy-read toggle, voice)
- Coverage: # LGAs & service categories supported
- Human handover outcomes: % resolved within 3 business days
This project demonstrates how agentic AI can simplify access to essential services, strengthen trust, and promote inclusion.
CommunityMate isn’t just about information—it empowers residents to act, reduces barriers to vital support, and fosters stronger community engagement.
Try the Live Demo
Click here to launch the CommunityMate Demo
