Autogeneration of Support Responses: LLM and Agent Assist

A support operator handles 500 similar requests per day. Average response time is 4 hours, CSAT is dropping, and the team is burning out. LLM autogeneration of responses solves this: we implement a system that either fully answers standard questions or suggests a ready response to the operator. The second mode—agent assist—provides a 40-60% reduction in response time without losing quality. According to a study by Gartner, companies that implemented agent assist reduced request processing time by an average of 45%. With a flow of 500 requests per day, payroll savings can reach up to 180,000 rubles per month (10 operators with an average salary of 60,000 rubles each).

We have completed 30+ projects on autogeneration for banks, retail, and telecom. We guarantee: acceptance rate >60%, response time reduction from 40%, and CSAT not lower than manual responses.

How does agent assist work?

We use a RAG architecture: the LLM (GPT-4o, LLaMA 3, or Mistral) is augmented with a vector knowledge base (Pinecone, Qdrant) containing historical resolved tickets and articles. The system operates in two modes:

• Fully automatic responses — for requests with a classifier confidence >0.95. Safe only for trivial cases: order status, working hours, confirmations.
• Agent Assist — the LLM generates a suggestion, and the operator accepts, edits, or rejects it with one click. Suitable for 70-80% of inquiries, including complex ones.

Agent assist is more effective than full automation because the operator stays in the loop: they check facts, add empathy, and adjust tone. Full automation without control carries a risk of reputational loss due to hallucinations. Agent assist provides 40-60% time savings with a CSAT level matching manual responses.

Example code: Agent Assist in Python

def suggest_response(ticket: Ticket, knowledge_base: VectorStore) -> ResponseSuggestion:
    # Search for similar resolved tickets
    similar_tickets = knowledge_base.search(ticket.text, top_k=3)

    # Search in knowledge base
    kb_articles = knowledge_base.search_articles(ticket.text, top_k=3)

    # Generate suggestion
    prompt = f"""
Customer inquiry: {ticket.text}
Customer history: {ticket.customer_history}

Similar resolved inquiries:
{format_similar(similar_tickets)}

Knowledge base articles:
{format_articles(kb_articles)}

Write an operator response: polite, to the point, with a concrete solution."""
    return llm.generate(prompt)

Comparison of modes: automatic vs assistant

Metrics comparison before and after implementation

Why RAG is better than direct LLM call?

Without RAG, the LLM generates a response based only on training data—this gives a high risk of hallucinations on company-specific questions. RAG adds context: it retrieves relevant documents from the knowledge base and passes them into the prompt. This ensures the response is based on facts, not generalities. We use vectorization (1536-dim) and re-ranking to increase accuracy.

How we implement autogeneration: stages

1. Analysis of the inquiry flow — collect statistics: top 20 topics, language, complexity. Determine which requests can be automated. Use a BERT-based classifier.
2. Collecting and preparing the knowledge base — consolidate ticket history, FAQ, articles. Clean, label, create embeddings (1536-dim). Typical base size is 5000+ documents.
3. Configuring the RAG pipeline — choose a model (GPT-4o for SaaS, LLaMA 3 for on-prem), configure the retriever (top_k=5), prompt with instructions. Add relevance ranking.
4. A/B testing — run agent assist on 10% of flow, compare with control group. Metrics: response time, CSAT, acceptance rate (target >60%). After 2 weeks, scale up.
5. Deployment and monitoring — deploy in Docker/Kubernetes, connect monitoring (p99 latency, FLOPS, GPU utilization). Set alerts for quality drops.

What is included in the work

• Architectural document — solution description, model selection, integration scheme.
• Integration with CRM/ticketing system — REST API, ready connectors to Zendesk, Freshdesk, Bitrix24.
• Operator training — 2-hour session: how to work with hints, what to do in case of errors.
• Monitoring and SLA — Grafana dashboard, weekly reports. Support for 2 months after implementation.
• Model training on your data — fine-tuning for increased accuracy (optional).

Why choose us

We are a team of AI/ML engineers with experience in production systems. We have completed 30+ projects on autogeneration for banks, retail, and telecom. We guarantee: acceptance rate >60%, response time reduction from 40%, CSAT not lower than manual responses.

Contact us for a detailed business case with ROI calculation. Order a pilot — we will analyze your flow and propose an architecture in 2 weeks. Pilot project cost starts at $5,000, covering analysis and a working prototype.

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.