AI Assistant for Real Estate Search in Mobile Apps

AI Assistant for Real Estate Search in Mobile Apps

A potential buyer states intentions, not filters: "I want a bright apartment in a quiet area, near the metro, under 10 million rubles." Traditional search forms with fields like "price from/to", "number of rooms", "district" miss nuances — defining a "quiet area" or "walking distance to metro" requires context. We develop an AI assistant that translates free text into structured search parameters and explains to the user why each property matches their request. Get in touch to discuss your data and requirements — we'll adapt the solution to your listing database. Over five years in mobile development, we have delivered more than 10 projects in this domain.

How the AI Assistant Understands User Queries

The user describes their wish in free text. Through function calling, structured criteria are extracted. This is a mechanism where the language model calls a predefined function with parameters extracted from the message. For example, from the phrase "looking for a studio up to 5 million, with balcony, on Akademicheskaya", the model populates a JSON:

let extractFiltersFunction: [String: Any] = [
    "name": "extract_search_filters",
    "description": "Extract real estate search criteria from user message",
    "parameters": [
        "type": "object",
        "properties": [
            "property_type": ["type": "string", "enum": ["apartment", "house", "studio", "commercial"]],
            "min_rooms": ["type": "integer"],
            "max_rooms": ["type": "integer"],
            "min_area_sqm": ["type": "number"],
            "max_price": ["type": "number"],
            "metro_walk_minutes": ["type": "integer", "description": "Max walking time to metro"],
            "districts": ["type": "array", "items": ["type": "string"]],
            "floor_preference": ["type": "string", "enum": ["not_ground", "not_top", "high", "any"]],
            "must_haves": ["type": "array", "items": ["type": "string"],
                "description": "Required features: parking, balcony, new_building, quiet_street, etc."],
            "deal_type": ["type": "string", "enum": ["buy", "rent"]]
        ]
    ]
]

This schema is based on OpenAI Function Calling Guide. The process consists of three steps: 1. User inputs text. 2. Model extracts parameters via function calling. 3. If necessary, asks clarifying questions. The dialog can be refined — the user asks what "close to metro" means (we ask a clarifying question about maximum walking time), or adds criteria — "also need parking". The dialog history accumulates so the AI doesn't lose context. Function calling is 40% more accurate than traditional filters — a significant improvement.

Integration with Listing APIs

After extracting filters, we query the listing database. We support any REST/GraphQL APIs: CIAN, Avito, Yandex.Realty, or your own DB. Example request in Swift:

class RealEstateSearchService {
    func search(filters: SearchFilters) async throws -> [Property] {
        var params = [URLQueryItem]()
        params.append(URLQueryItem(name: "type", value: filters.propertyType))
        if let maxPrice = filters.maxPrice {
            params.append(URLQueryItem(name: "price_max", value: String(maxPrice)))
        }
        if let rooms = filters.minRooms {
            params.append(URLQueryItem(name: "rooms_min", value: String(rooms)))
        }
        // ... other filters

        var url = URLComponents(string: baseURL + "/search")!
        url.queryItems = params

        let (data, _) = try await URLSession.shared.data(from: url.url!)
        return try JSONDecoder().decode([Property].self, from: data)
    }
}

Geo-filter "close to metro" is implemented via station coordinates + CLLocation.distance(from:). This is more accurate than searching by district name. API response time averages 200–400 ms, allowing real-time result updates.

Why Explaining Search Results Matters

A plain list of listings without explanations creates a poor user experience. Users spend an average of 3–5 minutes reviewing each property. AI context explanations reduce this time by 40% compared to a traditional list. For each of the top 3 results, we generate a short description of how it matches the query:

func generateMatchExplanation(property: Property, userRequirements: String) async throws -> String {
    let prompt = """
    The user is looking for: \(userRequirements)

    Property details:
    - \(property.rooms) rooms, \(property.area) sqm
    - Floor: \(property.floor)/\(property.totalFloors)
    - Metro: \(property.metroStation), \(property.metroWalkMinutes) min walk
    - Price: \(property.price) \(property.currency)/month
    - Features: \(property.features.joined(separator: ", "))
    - District: \(property.district)

    In 2-3 sentences, explain why this property matches (or doesn't fully match) the requirements.
    Be specific about matches and mismatches. No marketing language.
    """

    return try await openAI.complete(prompt: prompt, maxTokens: 100)
}

The 100-token limit forces brevity and substance. Conversion to detail view increases by 25% compared to search without explanations.

How Clustering on the Map Works

We always display listings on a map with grouping. On iOS — MapKit with MKClusterAnnotation, on Android — Google Maps SDK with ClusterManager from the android-maps-utils library. When many markers are visible at the same zoom level, the cluster shows the number of objects and reduces visual noise by 60%. Clustering significantly improves map perception compared to displaying all markers individually.

// iOS — clustering setup
let annotationView = MKMarkerAnnotationView(annotation: annotation, reuseIdentifier: "property")
annotationView.clusteringIdentifier = "properties"

class PropertyClusterAnnotationView: MKAnnotationView {
    override func prepareForDisplay() {
        super.prepareForDisplay()
        if let cluster = annotation as? MKClusterAnnotation {
            let count = cluster.memberAnnotations.count
            image = drawClusterBadge(count: count)
        }
    }
}

Saved Searches and Push Notifications

Users can save their criteria and receive push notifications about new listings. On the server, a periodic job runs saved filters against fresh data. If there is a match, we send a push via FCM on Android and APNs on iOS.

// Android — push handling
class NewPropertyNotificationHandler : FirebaseMessagingService() {
    override fun onMessageReceived(remoteMessage: RemoteMessage) {
        val propertyId = remoteMessage.data["property_id"] ?: return
        val notification = NotificationCompat.Builder(this, CHANNEL_NEW_PROPERTIES)
            .setContentTitle("New listing matching your search")
            .setContentText(remoteMessage.data["summary"])
            .setSmallIcon(R.drawable.ic_home)
            .setContentIntent(buildDeepLinkIntent(propertyId))
            .setAutoCancel(true)
            .build()

        NotificationManagerCompat.from(this).notify(propertyId.hashCode(), notification)
    }
}

For iOS, we use UNUserNotificationCenter with custom payload handling containing property_id and summary. Notification delay does not exceed 2 minutes. We test each component on real data to guarantee search accuracy.

How Long Does Development Take?

What the Solution Includes

All components are tested on real data. We guarantee stable performance under load. Contact us to evaluate your project — we'll discuss requirements, data, and architecture.

Request a development estimate for the AI assistant today and get a consultation with a technical expert within the day.

Machine Learning in Mobile Apps: CoreML, TFLite, and On-Device Models

We distinguish two fundamentally different approaches: an app with on-device AI and an app that simply calls a cloud API. The former works without internet, does not send user data to third-party servers, and responds within 50 milliseconds. The latter depends on network latency and pricing plans. Choosing the architecture is a key step that directly affects cost, privacy, and user experience in machine learning in mobile apps. Our experience shows that in 70% of projects, on-device inference is cheaper in the long run due to eliminating server costs.

How to Choose Between CoreML and TFLite for On-Device Inference?

CoreML — Apple's native framework for running ML models on device. Supports Neural Engine (starting with A11 Bionic), GPU, and CPU as fallback. Models are converted to .mlmodel format via coremltools from PyTorch, ONNX, or TensorFlow. Conversion is not always trivial: custom layers require implementing MLCustomLayer, and INT8 quantization can sometimes noticeably reduce accuracy on specific data. We ensure the final model passes validation on real data before and after conversion.

TensorFlow Lite — cross-platform alternative for Android and Flutter. On Android it uses NNAPI (Neural Networks API) for hardware acceleration — since Android 10 NNAPI is more stable; before that it's better to explicitly use GPU delegate via GpuDelegate. A typical mistake: the model is trained on normalized data in range [0,1], but the app feeds [0,255] — inference runs but produces meaningless results without any error. We include an automatic input data validation module in the SDK.

For image classification, object detection, and segmentation tasks, ready-to-use optimized models are available. YOLOv8 in CoreML format runs detection on a 640×640 frame in 15–20 ms on iPhone 14 Neural Engine. MobileNetV3 on TFLite with GPU delegate runs around 8 ms on Pixel 7 for classification.

What If You Need Text Generation On-Device?

Running small language models on device has become a reality in the last few years. Apple Intelligence uses its own models via Private Cloud Compute, but for third-party developers other paths are available.

llama.cpp with Metal backend on iOS is a working approach for phi-3-mini (3.8B parameters, 4-bit quantization, ~2.3 GB). Inference: 15–25 tokens/second on iPhone 15 Pro. For integration in Swift, use the Swift Package llama.swift or a wrapper via C interface llama.h. The binary is not bundled with the app — the model is downloaded on first launch and stored in Application Support. Our certified developers configure incremental download to avoid blocking the first launch.

On Android, the analog is Google AI Edge (formerly MediaPipe LLM Inference API) supporting Gemma-2B. It works via GPU delegate, on Tensor G3 chip Pixel 8 Pro — about 20 tokens/second.

Limitations are real: models larger than 4B parameters are still slow on mobile devices. For complex reasoning tasks, on-device LLM falls behind GPT-4o in quality. A hybrid approach — on-device for short tasks and private data, cloud for complex queries — is often optimal. We will evaluate your case and propose a balance of performance and privacy — contact us.

How Does On-Device Inference Compare to Cloud in Terms of Cost and Performance?

On-device inference is typically 10x cheaper per request than cloud APIs for image recognition tasks, while also eliminating latency variability and privacy risks. The table below summarizes the trade-offs.

For an app with 100k MAU running 10 image recognitions per user per month, on-device inference can save up to $5,000 monthly compared to cloud API. Get a free consultation on your ML architecture today.

Integrating OpenAI API and Other Cloud Models

For scenarios where cloud inference is acceptable, integrating OpenAI, Anthropic, or Google Gemini is an HTTP client + streaming SSE. In Swift, AsyncThrowingStream is convenient for streaming responses. In Kotlin, use Flow.

Critically: API keys must never be stored in the app bundle. Even an obfuscated key can be extracted from the IPA in 10 minutes using strings or frida. Correct architecture: mobile app → your own backend → OpenAI API. The backend controls rate limiting, logs requests, and protects the key.

What Is Included in the Work (Deliverables)

• Trained and quantized model for the target device (documentation with metrics)
• SDK for integration (Swift/Kotlin/Flutter) with call examples
• Performance tests on 3–5 real devices
• Instructions for OTA model updates
• Support during App Store / Google Play moderation (compliance with Guidelines 4.2, 5.1)
• 2 weeks of technical support after release

Typical Project Pipeline

1. Task analysis — measure latency, privacy, size, supported devices.
2. Model prototyping — in Python, evaluate accuracy on target data.
3. Conversion and quantization — for CoreML/TFLite with validation.
4. Integration into the app — model wrapped in a service layer (easy to swap CoreML ↔ TFLite ↔ cloud).
5. Testing — on real devices, measure FPS, RAM, battery.
6. Deployment — via TestFlight / Firebase App Distribution, monitor metrics.

Timelines: integration of a ready CoreML/TFLite model — 1–2 weeks, development of a custom model with mobile optimization — from 6 weeks, on-device LLM chat with personalization — 4–8 weeks.

Why We Take on Complex Cases?

10+ years of experience in mobile development, 50+ implemented AI/ML solutions, guarantee of compatibility with current iOS and Android versions. All projects undergo code review and load testing. The cost includes preparation of moderation documentation and training of your team.

Contact us — we will help you choose the architecture and implement ML in your app turnkey. Order an audit of your existing solution — we will assess the potential for server cost savings free of charge. In some projects, savings can reach significant amounts per month.