AI agent that finds, negotiates with, and hires a grass cutting service — entirely on your behalf via WhatsApp
Challenge: Hiring a grass cutting service in Malaysia involves multiple rounds of WhatsApp messages — finding providers, confirming coverage, negotiating rates, sharing the work area, confirming the booking, and verifying the job when it is done. This takes hours of back-and-forth communication that most property owners would rather not deal with.
Solution: Hire Gardener is an AI agent that handles the entire vendor communication workflow on your behalf. You tell it where the grass needs cutting, upload a photo and video of the area, and the AI does everything else — contacting providers, collecting quotes, confirming the booking, and verifying job completion. Service providers interact naturally over WhatsApp and never know they are talking to an AI.
AI identifies 3 grass cutting providers operating in the user's specified area.
AI sends a WhatsApp message to all 3 providers asking if they cover the target area.
For providers who respond positively, the AI sends the user's area photo and video, then requests a rate and available schedule.
AI compiles a quote comparison table. User selects their preferred provider.
AI contacts the chosen provider to confirm the rate, service date, and full job scope.
Provider sends completion photos. AI compares evidence against the agreed job scope and reports to the user. User approves payment.
A single config flag switches between full simulation (Llama3 plays all vendor roles) and live mode (real WhatsApp Cloud API). No code changes needed — just update your .env file.
A fully interactive HTML demo simulates the entire 6-stage workflow with authentic WhatsApp UI — green chat bubbles, typing indicators, media cards, and real-time message flow.
In mock mode, Llama3 plays three distinct vendor personas — a friendly Malay-English contractor, a professional English-speaking service, and one who declines mid-conversation for realism.
After collecting responses, the AI extracts rate and availability from each vendor and presents a clean comparison table for the user to make an informed decision.
User uploads an area photo and video before starting. These are sent to vendors during the quoting stage. In real mode, files are transmitted via the WhatsApp Cloud API.
When the provider reports job done, the AI compares their completion report against the original confirmed job scope and produces a structured verification report.
The AI agent operates with a system prompt that instructs it to speak naturally in Malaysian English/Malay, keep messages short and conversational, and never reveal it is an AI.
Each mock vendor has its own system prompt defining personality, pricing range, availability, and whether they decline the job — creating realistic, varied responses.
Each of the 6 workflow stages has a dedicated prompt engineered for that context — outreach, quote request, confirmation, completion report, and verification all use different instructions.
The messaging module is fully decoupled from the workflow logic. Swapping Llama3 for a cloud model (GPT, Claude) requires changing only one module — the workflow and UI remain unchanged.
config.py reads a single MODE env var — mock or real — and the messaging module loads accordinglymessaging/mock.py and messaging/whatsapp.py share the same function signature — swap with zero workflow changesollama pull llama3MODE=real and add your API credentials to .env