Pumpkin AI Agent Platform
The Pumpkin AI Agent Platform is a conversational AI agent platform deployed inside venues, exhibition floors, and enterprise sites. It comprises three modules — AI Guide, AI Instructor, and AI Concierge — answers from the knowledge base you supply, and deploys anywhere from cloud to fully offline.
- Modules
- AI Guide、AI Instructor、AI Concierge
- Answers from
- Your own knowledge base
- Deployment
- Cloud / private / on-premise / offline
- Starts with
- A proof of concept
Premise: every answer must be traceable
The platform is built on one premise: every answer must be traceable. Responses are confined to the scripts, standards, and procedures you supply rather than left to the model’s discretion — because on an exhibition floor, in a classroom, or at a front desk, the cost of one wrong sentence lands on you.
Three modules
AI Guide
The visitor-facing interpreter. It answers questions about exhibits, spaces, and themes, strings together tour routes, and unfolds deeper background layer by layer as visitors dig in. Unlike fixed audio guides it handles unscripted questions, and adjusts depth to a visitor’s age and background.
- Exhibit and theme Q&A
- Route recommendation and flow sequencing
- Tiered depth (children / general / specialist)
- Instant language switching
- Visitor interest and popular-question analytics
AI Instructor
The staff-facing trainer. Used for onboarding new staff and docents, rehearsing standard content and procedures, and querying regulatory documents. In instructor mode it poses questions, records answers, and points out gaps rather than only responding passively — so it doubles as a training record.
- Standard content and SOP rehearsal
- Question prompts, answer logs, and gap analysis
- Instant lookup of standards and operating documents
- Onboarding progress tracking
- Integrates with existing training records
AI Concierge
The entrance, front-desk, and booth agent. It handles visitor enquiries, wayfinding and facility guidance, session and ticketing questions, and group reservations. On a show floor it serves multiple visitors at once, absorbs repetitive questions, and captures contact details so your team can focus on real conversations.
- Visitor enquiries and FAQ handling
- Wayfinding, facilities, and accessibility guidance
- Sessions, ticketing, and opening hours
- Group reservations and requirement intake
- Trade-show lead and requirement capture
Integration interfaces
- Knowledge Base & Retrieval-Augmented Generation
- Your existing scripts, catalogues, standards, and FAQs are structured into a knowledge base. Answers retrieve first and generate second, with sources retained — this is how traceability is implemented: when you need to verify a sentence, you can find the passage it drew from.
- Content Management Back Office
- Your staff update exhibit descriptions, special-exhibition content, and FAQs themselves — no engineering involvement, no redeployment. Includes a pre-publish review flow and side-by-side proofreading across languages.
- On-Site Hardware Integration
- Runs on visitors’ own phones (scan to start), touch displays and standing kiosks, handheld guides and tablets, and AR / VR headsets. It rides on the hardware you already own — adopting the system does not require a hardware refresh.
- Existing System Integration
- Integrates with ticketing, membership, booking, and internal knowledge systems so answers can reference live data — today’s sessions, remaining capacity — rather than static descriptions. Scope and interfaces are confirmed during the PoC.
From PoC to production
- 01
Discovery & Site Assessment
1–2 weeksConfirm use cases, visitor mix, language requirements, and security constraints; inventory the completeness and digitization state of existing interpretation content.
DELIVERABLE Scope document and deployment recommendation
- 02
Proof of Concept
4–6 weeksBuild a usable prototype for one gallery or theme set, put it in front of real visitors on site, and score it against a jointly agreed test set.
DELIVERABLE Working on-site prototype and test-set scoring report
- 03
Production Build & Customization
2–4 monthsExpand to the full-venue knowledge base; complete interface and persona customization, multilingual proofreading, hardware integration, and existing-system interfaces; deploy into the designated environment.
DELIVERABLE Production system, admin console, and deployment documentation
- 04
Acceptance & Training
2–4 weeksAccept against the contracted criteria — correct-answer rate, response time, per-language proofreading pass rate, system availability — and train your staff on the admin console.
DELIVERABLE Acceptance report, operations manual, and training records
- 05
Maintenance & Content Updates
OngoingEither your IT team operates it with a technical support channel open on our side, or we take an annual maintenance contract covering system updates, model tuning, and troubleshooting.
DELIVERABLE Maintenance records and periodic performance review
Boundaries and acceptance
- Source of answers
- Confined to the scripts, standards, and FAQs you supply. The system does not answer from the open web, and every sentence can be traced back to the passage it drew from.
- What is customizable
- Interface visuals, persona and tone, visitor flow, knowledge-base structure, and integration with existing ticketing, CMS, booking, and membership systems. The persona can reuse your existing mascot or be designed fresh.
- What is not customizable
- The underlying capability of the language model itself. Our approach is to constrain it through the knowledge base and answering rules so it answers only from your material — not to claim the model never errs.
- Acceptance criteria
- Agreed with you case by case and written into the contract. Common items are answer accuracy, offline readiness, per-language proofreading, system availability, and handover — with threshold figures set by you to fit your site.
- Data residency
- Determined by the deployment method. With private cloud, on-premise, or offline deployment, the knowledge base and interaction records stay inside the environment you designate, with nothing sent overseas; the actual location, retention period, and access controls are confirmed in writing before the build.
The full acceptance protocol for guide deployments — a four-column table you can lift into a requirement document — is on the AI guide solution page.
Plans & Pricing
Quoted against venue scale, number of languages, module mix, and deployment method. After a discovery session we provide an itemized quote covering knowledge-base construction, customization, deployment environment, and maintenance — in a form suitable for internal approval and procurement paperwork.
Want to confirm it fits your site first?
Tell us the venue type, the systems you already run, and your security constraints. We will provide an architecture description, an integration recommendation, and a draft of acceptance clauses ready for contract.