AI Chatbot Development for Viber: Sales & Support Automation with NLP

Your client writes via Viber — a standard scenario for e-commerce in the CIS. Manual support of 1000+ dialogues a day eats the budget and reduces conversion: the average operator response time is 4 minutes, while a bot processes a request in 2 seconds. Developing an AI chatbot with NLP and integration through the Viber API turns the messenger into an automated sales and support channel 24/7. We use a production-ready stack: from GPT-4o to locally deployed models via vLLM with INT8 quantization to reduce latency p99 to 1.5 seconds. In one project, the conversion to order increased by 18% after implementing a product carousel — Rich Media allows selecting products right in the chat.

What tasks an AI bot for Viber solves

Inbound processing with NLP. The bot understands natural language, distinguishes a complaint from a delivery question, and switches to an operator on escalation. With few-shot prompts, classification accuracy reaches 92–95% within the first weeks of operation. Additionally, we configure intent detection with a threshold of 0.85 — if confidence is lower, the dialogue is handed over to a human.

Rich Media and catalogs. Viber supports cards with buttons and images. For online stores, this means: the user selects a product right in the chat — without going to the website. We implement dynamic rich_media responses based on data from your CRM.

How the AI model processes dialogues in real time?

After registering the bot through the Viber Partner Program, we configure a webhook to an HTTPS endpoint. All incoming messages go into a queue (via Redis), then are passed to a LangChain agent with dialogue history. A model router decides: answer from FAQ (RAG) or pass to the LLM. The average response delay is 1.5–3 seconds, which fits within messenger UX expectations. For critical scenarios, we use chain-of-thought prompts with 5-shot examples.

LangChain is more efficient than pure REST requests: it provides calls to external tools (database search, order status check) and context management. In our tests, this approach reduces false positives by 30% compared to direct API calls to the model.

How to choose an LLM for a Viber bot?

Choosing a model is a trade-off between quality and speed. The table compares popular options.

For simple FAQs, Mistral 7B is enough. If you need RAG with deep reasoning, LLaMA 3 70B in INT8 provides balance. For maximum quality, we stick with GPT-4o, but with caching of frequent requests.

Comparison of RAG and fine-tuning

Development process: from analysis to deployment

1. Analysis. We collect typical scenarios, define metrics (CSAT, FCR, conversion to order). We analyze 2–3 months of support logs.
2. Design. We choose the LLM, vector DB (Qdrant), design dialogue graphs. We define the fallback strategy to a human operator.
3. Implementation. We write a webhook handler in Python (FastAPI), integrate the Viber API, connect the RAG pipeline with chunk size 512 tokens.
4. NLP testing. We run 200+ test dialogues, measure accuracy, latency p99, hallucination rate. We tune temperature (0.1–0.3) and top-p.
5. Deployment and monitoring. We deploy on Kubernetes (SageMaker or Vertex AI), set up logging via MLflow and alerts in Grafana. We establish an SLA of 99.9%.

What is included in the result

• Documentation on the architecture and API of all components.
• Access to the admin panel for managing scenarios and viewing dialogue logs.
• Operation manual for the business customer.
• 2 weeks of post-launch support (bug fixes, prompt adjustments).
• Certified ML engineers ensure monitoring and model fine-tuning on your data.

Why trust the integration to our team

7+ years of experience developing AI solutions for messengers and web channels. 50+ successful integrations with Viber, Telegram, and WhatsApp. We guarantee migration to your model (LLaMA, Mistral) without quality loss. Every project undergoes MLOps audit — we don't deploy black boxes into production.

Contact us for a consultation — we will select the architecture for your budget and timeline. Get a project estimate: an engineer will analyze your scenarios and propose the optimal solution. Order a prototype of an AI bot for Viber today — just write to us.

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.