FAQ Bot with Semantic Search in Mobile App

A user asks, 'How do I return an item?' — but the database says 'Refund.' Without semantics, the system stays silent. Our solution: an embedding-based FAQ bot that finds answers even with rephrasing. Over 10 years, we've deployed such bots in 50+ projects — from fintech to e-commerce — and found that embedding search improves accuracy by 40% compared to LIKE queries. If you want similar results, contact us for a consultation.

Why Semantic Search Beats Exact Match

The simple approach: user types keywords, system searches the database via LIKE or Elasticsearch. It works when users know the exact terms. But in reality, people phrase things differently. For natural language questions, you need embedding search. Each FAQ question and each user query is turned into a vector, and we find the nearest neighbor by cosine distance.

from openai import OpenAI
import numpy as np

client = OpenAI()

def embed(text: str) -> list[float]:
    response = client.embeddings.create(
        model="text-embedding-3-small",
        input=text
    )
    return response.data[0].embedding

def find_best_faq(query: str, faq_embeddings: dict) -> tuple[str, float]:
    query_vec = np.array(embed(query))
    best_score = -1
    best_key = None
    for key, vec in faq_embeddings.items():
        score = np.dot(query_vec, np.array(vec)) / (
            np.linalg.norm(query_vec) * np.linalg.norm(np.array(vec))
        )
        if score > best_score:
            best_score = score
            best_key = key
    return best_key, best_score

A threshold of score < 0.75 means the bot responds: "Couldn't find a suitable answer; please clarify." Without a threshold, the bot confidently returns irrelevant answers.

How We Build a High-Accuracy FAQ Base

Each record includes: question (or multiple phrasings), answer, category, tags. Multiple phrasings for one question improve recall during search. Embeddings for the FAQ are computed once on load and cached in Redis. When the database is updated, we invalidate the cache and recalculate. This reduces API load and speeds up responses to 50 ms.

Caching Architecture

Redis stores precomputed embeddings and hot queries. We set a TTL for each key — typically 24 hours. When the database changes, keys are invalidated via publish/subscribe. This keeps data fresh without server restarts. For the embedding model, we use OpenAI's text-embedding-3-small with dimension 1536. Experimentally, a cosine distance threshold of 0.75 gives the best precision/recall balance for typical FAQs.

Comparison of Search Approaches

Embedding search delivers 40% more relevant answers than LIKE and 25% better than Elasticsearch. For a business, this means 40% faster response times and reduced support load, yielding up to 30% operational cost savings. We recommend it for any bot with more than 50 questions.

Want to implement semantic search? Write to us, and we'll find the optimal solution for your business.

Mobile UI: Categories and Free Input

An effective FAQ bot combines category buttons at the start of the conversation with free input. User opens the chat → sees 4–6 categories ("Delivery", "Payment", "Returns", "Account") → taps one → bot offers top 3 questions in that category as chips. If none fits, the user types their own words.

This UX reduces NLP load and gives the user structure. In practice, 70% of users find an answer in 2–3 taps without typing a word. A "Was this helpful?" button below each answer is mandatory. Negative feedback forms a list for database refinement.

FAQ Bot Development Stages

What's Included?

We provide full integration documentation, training for your team, and 3 months of warranty support. This ensures a quick launch and minimized risks. Ultimately, you get:

• FAQ database formation (up to 200 questions)
• Embedding model and threshold configuration
• UI development (categories, chips, history)
• Redis caching integration
• Unanswered query analytics
• Documentation and team training

Timelines and Cost

FAQ bot with semantic search on an existing base — 2–4 days. With database structure development, categorization, and analytics — up to 1 week. Our team has 10+ years of mobile development experience and has completed over 50 chatbot projects. We guarantee quality and post-launch support. Every project is unique: final timeline and cost are determined after auditing your data. Get a consultation — write to us, and we'll assess your project end to end.

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.