Alquilia Case Study: An AI Assistant That Qualifies Rental Tenants on WhatsApp

Alquilia is an AI-powered platform that automates how real estate agencies capture, qualify and screen rental tenants. Instead of agents spending hours chasing leads and chasing paperwork, an intelligent assistant handles the entire first stage of the funnel: from the moment an inquiry lands to a fully scored, document-backed candidate ready for a human decision. In this case study we walk through the problem the client brought us, the solution we designed, how the AI actually works under the hood, and the engineering behind a production-grade, multi-tenant SaaS.

The Challenge

Rental agencies in Spain live and die by speed. Studies of portals like Idealista show that the agent who replies first wins the lead the vast majority of the time, yet a human team can only answer so fast, and only during working hours. On top of that, every inquiry has to be qualified (income, employment, solvency, occupancy, move-in date) and backed by a stack of documents (ID, payslips, tax returns, bank statements, contracts). Done manually, this is 20+ hours a week per agent of repetitive screening, and most of those leads turn out to be unqualified anyway. The client needed to respond instantly, around the clock, filter out bad fits automatically, and only hand real, qualified candidates to their team.

Our Solution

We designed and built Alquilia end to end: a SaaS platform whose core is an AI assistant that talks to every incoming lead on WhatsApp (the channel Spaniards actually use) within seconds, 24/7. The assistant introduces the property, runs a structured qualification conversation, requests and reads the required documents, calculates an objective risk score, and either advances the candidate to the agency or politely closes the conversation. Around that core we built everything an agency needs to run on it: multi-tenant dashboards, configurable AI behaviour, billing, analytics, portal and email integrations, and full GDPR compliance for automated tenant evaluation.

How It Works: End to End

A single lead flows through the system automatically, with a human stepping in only at the very end:

• 1. Lead capture: Inquiries arrive from listing portals and agency inboxes. The system reads incoming portal emails, intelligently extracts the prospect's contact details and the property they're asking about, and creates the lead, matching it to the right listing automatically.
• 2. Instant WhatsApp contact: Within seconds the AI assistant reaches out on WhatsApp, introducing itself and the specific property with a personalized, on-brand message, with no human involved, day or night.
• 3. Conversational qualification: The assistant runs a natural, structured conversation to collect everything that matters: budget, income, employment, occupancy, pets, move-in date and more, adapting to the prospect's answers instead of reading a rigid form.
• 4. Document collection & reading: It then requests the right documents for each profile type, accepts photos, PDFs and even voice notes, and uses AI to read, extract and validate them, checking that they're real and consistent with what the prospect said.
• 5. Objective scoring: Using the agency's own rules, the system calculates a clear GREEN / YELLOW / RED score, with the reasons behind it, turning a subjective gut-feel into an auditable, consistent decision.
• 6. Human handoff: Only qualified, document-backed candidates are passed to the agency's team, with the full conversation and score attached. Bad fits are filtered out automatically and respectfully.

The AI Under the Hood

Alquilia isn't a thin wrapper around a chatbot. It's a layered AI system where each capability solves a distinct, hard problem, exactly the kind of end-to-end AI engineering we specialize in:

• Conversational engine with a real state machine: Rather than letting a model improvise, the assistant runs through explicit conversation stages (greeting, qualifying, documenting, scoring, handoff), each driven by AI but bounded by deterministic logic. The result is a bot that feels natural yet never goes off-script or skips a required step.
• Long-term conversation memory: The system maintains a persistent memory of each lead (confirmed facts, inferred details, and things the prospect said they'd send later), so it never re-asks a question and keeps full context even across days-long, interrupted conversations.
• Computer vision for documents: Uploaded IDs, payslips, contracts and bank statements are read with a vision model that extracts the relevant data and flags inconsistencies, turning a folder of images into structured, verifiable information.
• Speech-to-text for voice notes: Because real WhatsApp users send audio, the assistant transcribes Spanish voice notes automatically and treats them like any other message, with no dead ends in the conversation.
• Structured data extraction: Free-form chat is continuously distilled into a structured profile with confidence levels, so a messy human conversation becomes clean data the scoring engine and the agency can rely on.
• Vendor-agnostic LLM layer: Every model call sits behind a single abstraction with caching, rate limiting, circuit breakers and fallbacks, so the platform stays fast, cost-controlled and resilient even if a provider has a bad day.
• Human-like messaging: Responses are split into natural fragments with realistic typing delays and subtle variation, so prospects engage with an assistant that feels like a person, not a form.

Platform Capabilities

• Configurable AI personality: Each agency tunes the assistant's name, tone, emoji level, verbosity and how it introduces itself, so the bot speaks in the agency's voice, not a generic one.
• Agency-defined rules: Qualification questions, 'kill' criteria, required documents per profile type and scoring thresholds are all configurable per agency through a no-code dashboard.
• Multi-tenant SaaS: Agencies, users, properties, leads and subscriptions are fully isolated, with role-based access for owners, managers and agents.
• Real-time WhatsApp panel: Live conversations stream into the dashboard via server-sent events, with QR connection, message history and a built-in chatbot tester.
• Analytics & billing: Lead trends, conversion insights, source attribution and usage are tracked alongside Stripe-powered subscriptions, invoicing and pay-as-you-go lead packs.
• Admin control plane: A platform-level dashboard monitors agencies, subscriptions, AI provider health, security logs and conversation quality across the whole system.

What the Platform Delivers

• First response to every rental inquiry in seconds, 24/7, instead of hours and only during office hours
• Every lead pre-qualified and scored automatically before an agent ever gets involved
• Tenant documents requested, collected, read and validated without manual chasing
• Objective, consistent, auditable GREEN/YELLOW/RED decisions instead of subjective gut-feel
• Agents freed from repetitive screening to focus on the candidates that actually convert
• A multi-tenant product ready to scale across many agencies from a single codebase

Technology Stack

• React 19 + TypeScript + Vite for the frontend single-page app
• Java 17 + Spring WebFlux (fully reactive, non-blocking) for the backend
• PostgreSQL with reactive R2DBC access and versioned migrations
• Real-time updates via Server-Sent Events (SSE)
• Multi-model LLM orchestration: conversational AI, vision and speech-to-text
• WhatsApp Business integration as the primary conversational channel
• Stripe for subscriptions, invoicing and usage-based billing
• AWS S3 for secure document and invoice storage
• JWT auth with 2FA, Google/email OAuth, and encryption of sensitive data

Privacy & Compliance by Design

Automatically scoring people for housing is a high-stakes, regulated activity under GDPR. We built Alquilia with that front and center: a Data Protection Impact Assessment for the automated evaluation, records of processing activities, a security-incident protocol, configurable human-review for edge cases, encryption of sensitive data at rest, and full audit trails. Compliance isn't a page bolted on at the end; it's part of the architecture.

Engineering Challenges We Solved

• Making AI feel human, reliably: Balancing free-flowing conversation with the guarantee that every required step happens, solved with a hybrid of AI generation and a deterministic state machine, plus a humanization layer for natural pacing.
• Multi-modal, messy real-world input: Real prospects send blurry photos, PDFs and voice notes at random points in the chat. We built a pipeline that ingests any of them, reads them with the right model, and keeps the conversation coherent.
• Resilience and cost control at scale: Frequent AI calls can get slow and expensive fast. Caching, rate limiting, circuit breakers and a vendor-agnostic model layer keep the system responsive and predictable.
• Reactive, multi-tenant scale: A fully non-blocking backend with distributed scheduling lets a single deployment serve many agencies and thousands of concurrent conversations without blocking threads.
• Regulated automated decisions: Designing an auditable, configurable, human-reviewable scoring system that an agency can defend under GDPR, not a black box.

Why This Matters

Alquilia is proof that we don't just 'add a chatbot.' We design and ship complete AI products: conversational systems with memory, computer vision, speech-to-text, intelligent data extraction, automated decisioning and resilient multi-provider orchestration, all wrapped in a scalable, compliant, production-grade SaaS. Whatever the domain, if there's a real-world process that AI can capture, qualify or automate, we can build it end to end.

• Explore our AI agents & automation services
• Read the Almia AI platform case study
• Have a process you want to automate with AI?