AI Chatbot for Restaurants: Order Automation & Reservations

A restaurant loses up to 30% of incoming orders if a client waits for an operator's response longer than 20 seconds. We have encountered situations where the stop list changes every 15 minutes, yet the bot offers a dish that is already out of stock. Our AI chatbot solves both problems: it takes orders in 30 seconds and synchronizes dish status with the cash register system in real time. Instead of 3 minutes of waiting — 30 seconds, and no errors due to an outdated stop list.

The chatbot processes orders 6 times faster than an operator. Savings on personnel reach 60% — instead of 5 operators, 1 is enough, and call center costs are reduced by a factor of three. Payback occurs within 2–3 months. For a pilot project costing $2,500, clients save $12,000 monthly — a 5x ROI.

What problems do we solve

Loss of orders during peak load. At lunchtime, operators cannot handle all calls — average wait time exceeds 2 minutes. The chatbot processes 100% of requests instantly, using asynchronous models (e.g., GPT-4o with latency p99 < 1 s).

Errors in the stop list. If synchronization happens once an hour, the bot sells what the kitchen doesn't have. We configure WebHooks from iiko or R-Keeper — the stop list updates in seconds. Visitors don't face rejection after placing an order.

Low conversion from menu to order. The user views 10+ items and leaves. We implement RAG with a vector database (e.g., Qdrant) so the bot recommends dishes by parameters: 'something gluten-free', 'most popular hot dish'. Upselling desserts and drinks increases the average check by 15–20%.

Why is an AI chatbot faster than an operator?

The bot handles 80% of routine questions: taking orders, clarifying the address, delivery time, reservations. The operator only gets involved in complex cases (complaints, changing composition for an allergy sufferer). This reduces the need for a call center.

Source: internal metrics from 50+ implementations

Integration with your cash register system

We connect to iiko API, R-Keeper, Poster, 1C, or create a REST layer for systems without a public API. For each, we develop a data model: menu, stop list, orders, payments.

Example WebHook handler code from iiko (Python + LangChain):

from fastapi import FastAPI, Request
from langchain.memory import ConversationBufferMemory

app = FastAPI()
@app.post("/webhook/iiko/stoplist")
async def update_stoplist(request: Request):
    data = await request.json()
    # update vector DB: remove embeddings of finished dishes
    qdrant_client.delete(collection_name="menu",
                         filter=data["stopList"]["items"])
    return {"status": "ok"}

Supported POS systems:

How does the bot handle complex modifications?

For complex modifications like 'burger without onions but with double cheese', we use few-shot prompts with examples. The model is fine-tuned on your menu, allowing it to correctly interpret any combination. The bot clarifies missing details and confirms the order before sending it to the kitchen.

Typical risks and their elimination

The most common mistake is ignoring the stop list: the bot offers dishes that are not in the kitchen. We connect WebHook synchronization — delay less than a second.

Another problem is lack of context: the customer changes the order mid-dialogue — the bot loses the thread. We add ConversationBufferMemory with a limit of 200 tokens.

Our work process

1. Analytics — study current scenarios (order, reservation, delivery), collect dialogue examples, determine data volume.
2. Design — choose architecture (RAG + fine-tuning or ready-made LLM), design dialogue tree.
3. Development — write microservices: NLU (Hugging Face Transformers), POS integration, vector DB, WebHook handlers.
4. Testing — simulate 1000+ sessions with different scenarios (including edge cases: allergy, missing dish, address change).
5. Deployment and monitoring — launch on your server or cloud (AWS, GCP), configure dashboards (MLflow, Prometheus).

Estimated timelines

From 2 weeks to 2 months, depending on the number of POS systems, the need for model training on specific vocabulary, and the complexity of logistics integration. Cost is calculated individually — each project is unique.

Deliverables

• Full-featured chatbot with support for Telegram, WhatsApp, VK, and a website widget.
• Integration with your cash register system (real-time synchronization of stop list, menu, orders).
• NLU model training on your menu (at least 50 examples per dish).
• Delivery tracking module with proactive notifications.
• Analytics dashboard: conversion, average check, popular dishes, stop list.
• API documentation and personnel training (2–3 hours).
• Code warranty — 6 months of free technical support.

Our company has 5+ years of experience in AI and 50+ successful implementations in the restaurant industry. We handle AI chatbot restaurant, chatbot food delivery, iiko chatbot integration, restaurant order automation, NLP for restaurant, RAG menu bot, chatbot table reservation, delivery tracking bot, business lunch bot, artificial intelligence restaurant, NLU restaurant orders, and fine-tuning for restaurant-specific needs.

Get a free consultation — we will evaluate your project in 1 day. Order a pilot project with a result guarantee. Contact us to discuss your restaurant.

NLP Development: Text Classification, NER, Embeddings, and Information Extraction

We often receive a task: process 50,000 support tickets — currently all manual. Dataset — 3,000 labeled examples, 12 categories, imbalance: one category occupies 40% of the sample, three at 1-2% each. Baseline accuracy — 78%. Sounds decent until you look at recall for rare classes: 0.31, 0.44, 0.28. These classes — complaints and churn threats — are most important to the business.

This is a typical NLP development project. The problem is not the algorithm but that accuracy is the wrong metric. Our experience across 30+ projects shows: we start by analyzing business metrics and only then choose the model.

Why accuracy is not the right metric for rare classes?

Accuracy ignores imbalance. If the "churn" class appears in 2% of cases, the model can predict "all good" and get 98% accuracy — but the business loses clients. Solution: F1 macro (averaged over all classes) or weighted F1. For NER — strict entity F1 (exact matches only). We guarantee: after choosing the correct metric, model quality becomes measurable and predictable.

Text Classification: From BERT to Distillation

BERT-like models are the standard for classification. ruBERT-base or ruBERT-large from DeepPavlov for Russian. multilingual-e5-large — for multiple languages in one pipeline. XLM-RoBERTa-large — a strong multilingual backbone.

Fine-tuning for classification: add a classification head on top of the [CLS] token, train for 3-5 epochs with lr=2e-5, weight decay=0.01. For imbalance — weighted CrossEntropyLoss or focal loss with gamma=2.0. Contact us — we will show a code snippet.

Imbalance case study. Dataset — 3,000 examples, imbalance 1:20. Solution: class_weight via sklearn + CrossEntropyLoss. Additionally — augmentation of rare classes via backtranslation (ru→en→ru through MarianMT). Recall for rare classes rose from 0.31 to 0.67 with a slight drop in accuracy (76%→74%). Full NLP development end-to-end took 3 weeks.

Distillation for production. BERT-large gives F1 0.89, but inference on CPU — 180ms. Distillation into DistilBERT or ruBERT-tiny2 reduces latency to 25ms with F1 0.84. Export to ONNX Runtime provides an additional 1.5-2x speedup. DistilBERT achieves 7x lower latency than BERT-large with only a 5% drop in macro F1 – a typical production trade-off.

How to choose between BERT and LLM for your task?

For most classification and extraction tasks, BERT-sized models offer the best trade-off between cost and performance. Shift to LLMs only when the task demands generation, complex reasoning, or zero-shot generalization.

NER: Named Entity Recognition

NER — extracting persons, organizations, locations, dates, amounts, document numbers. For general categories (PER, ORG, LOC), pre-trained models work well. For specialized ones (medical terms, legal concepts) — fine-tuning is needed.

Data annotation. The main cost of an NER project. For a quality model — 500-2,000 labeled sentences per entity type. Tools: Label Studio (open source) or Prodigy (by spaCy creators). IOB2 format — standard.

Architecture. Token classification on top of BERT: each token gets a label (B-PER, I-PER, O). spaCy 3.x with transformer pipeline — a convenient production choice.

Nested entities. Standard IOB models cannot handle nested entities (organization inside an address). For such tasks — span-based NER: SpanBERT or SpERT. More complex but correct.

Post-processing is mandatory. The model predicts tokens — normalized entities are needed. Date — dateparser. Amounts — regex + validation. Names — deduplication via rapidfuzz. Included in our standard delivery.

Sentiment Analysis and Opinion Mining

Binary classification positive/negative works out of the box with BERT. Complexity — aspect-based sentiment analysis (ABSA): "the restaurant has good food but terrible service." For ABSA: aspect extraction (NER) + sentiment per aspect. Joint models BERT-for-ABSA — quality on Russian data is lower due to dataset scarcity. RuSentiment, SentiRuEval — main resources.

For production with simple positive/negative/neutral: distil models are enough. Three classes, balanced dataset, 2,000+ examples — F1 macro 0.82-0.87 in 1-2 days.

Text Summarization

Extractive summarization (select sentences) — TextRank or BM25 without training. Fast, no hallucinations. Good for long documents.

Abstractive (generates new text) — seq2seq: mT5, mBART, FRED-T5, ruT5-large. For production via LLM API (GPT-4, Claude) — often the best cost/quality/speed trade-off.

Embeddings: Vector Representations of Text

Embeddings are the foundation of semantic search, deduplication, clustering, RAG. Quality critically affects downstream tasks.

Models. E5-large-v2, BGE-M3, multilingual-e5-large — strong multilingual embedders. sentence-transformers/paraphrase-multilingual-mpnet-base-v2 — fast option. For Russian: ru-en-RoSBERTa (Skoltech) performs well on semantic textual similarity.

Embedding quality evaluation uses the MTEB benchmark as standard. But top results on MTEB don't guarantee success on a domain dataset — we build domain-specific eval.

Fine-tuning embeddings. If standard models don't give the required Recall@k — contrastive learning on domain pairs with MultipleNegativesRankingLoss. How to perform this for domain data:

1. Collect 500–2,000 semantically similar pairs from your domain.
2. Apply MultipleNegativesRankingLoss with a batch size of 32–64.
3. Train for 1–3 epochs using AdamW (lr=2e-5).
4. Evaluate Recall@k on a held-out domain test set.

This approach yields a 5–15% improvement in Recall@k in practice.

Dimensionality and storage. E5-large: 1024 dim, float32 — 4KB per vector. For 10M documents — 40GB. Quantization int8 reduces to 10GB. FAISS IVF_PQ — more compact but with losses. Included in our deployment recommendations.

Information Extraction

Structured extraction is a frequent task. Examples: key contract terms, technical characteristics, dates and amounts from invoices.

1. Regex + rule-based. For INN, OGRN, amounts, dates — more reliable than neural networks. No data required.
2. NER + post-processing. For variable formats.
3. LLM with structured output. GPT‑4 / Claude with JSON schema — for complex documents. Cost: minimal per document. For 10k+ documents/day — we calculate the economics.

We guarantee a hybrid: regex/NER for typical fields + LLM for edge cases. Our guarantee is backed by years of production experience and more than 30 projects.

Work Stages

Prototype on existing data — 1-3 weeks. Production system with CI/CD — 1.5–2.5 months. Cost is calculated individually — get a consultation for a project estimate.

What's Included

• Model and pipeline architecture documentation
• Access to the model via REST API (FastAPI + ONNX)
• Client team training (2-hour webinar + Q&A)
• Accuracy guarantee on the agreed test set
• Months of post-delivery support (bug fixes, adaptation to new data)

Our Experience

Years of NLP projects from classification to RAG systems. The team includes ML engineers experienced with Hugging Face, spaCy, LangChain, MLOps. We use vLLM, Kubeflow, Weights & Biases — a production stack, not toys. Contact us to evaluate your NLP project within two days — request a free consultation on your text processing pipeline.