RPA Bots with LLM: Processing Unstructured Data

Classic RPA automation tools — UiPath, Automation Anywhere, Blue Prism — handle structured data and deterministic scenarios well. The problem arises when unstructured text appears: emails, PDF scans, free forms, chats. Here, RPA without AI either needs rigid templates or breaks at the slightest deviation. Integrating LLM into the RPA pipeline closes this gap, and we offer a turnkey solution.

A typical scenario: incoming invoices from 50 different suppliers — each with its own structure. Manual processing takes 3–5 minutes per document. After implementing an LLM integration, time drops to 15–30 seconds, with key field extraction accuracy of 92–96%. Compared to traditional methods: the LLM approach is 4 times more efficient than template parsers and doesn't require retraining when formats change. Clients report average monthly savings of $15,000 after deployment. Order a pilot project — we'll evaluate LLM applicability on your documents in one week.

How does the RPA-LLM architecture look in production?

Not every process step needs a language model. A sensible architecture splits tasks: the RPA engine handles navigation, clicks, data transfer between systems. The LLM is plugged in selectively — where text understanding, entity extraction, or fuzzy decision-making is needed.

Typical integration points:

• Data extraction from incoming emails — determining request type, extracting details, routing
• Processing PDF documents — invoices, acts, contracts with variable structure
• Classification of inquiries — support, complaints, information requests
• Form filling based on free-text descriptions or documents

The standard scheme includes three layers:

RPA Layer — process orchestrator. Depending on the platform, this could be UiPath Orchestrator, Robocorp, n8n, or a custom Python scheduler. Responsible for triggers, task queues, and result logging.

AI Processing Layer — a microservice or lambda that accepts unstructured content and returns structured JSON. Internally: text preprocessing (pytesseract/pdfminer for extraction, langchain/llama-index for orchestrating LLM requests). The model — GPT-4o, Claude 3.5 Sonnet, or local Mistral/LLaMA via Ollama, depending on confidentiality requirements.

Validation Layer — checks model confidence, falls back to human when confidence is low. Implemented via structured output (JSON Schema in prompt or OpenAI function calling) plus post-processing rules.

What's included in the work

• Architecture documentation and API specifications
• LLM microservice access via REST API
• Training for RPA developers
• One month of support after launch

Why confidence routing is critical for production?

The model is not always certain. Confidence routing strategy:

• confidence > 0.9 — automated processing, logging
• 0.7–0.9 — processing plus flag for selective review
• < 0.7 — send to manual review queue plus notification

Confidence can be obtained in several ways: token logprobabilities (available via OpenAI API), a separate verification prompt, or an ensemble of two models with voting. Our confidence routing architecture reduces human escalation by 80% compared to threshold rules.

Which LLMs are best for RPA?

Model choice depends on latency, accuracy, and confidentiality requirements. Typical LLM call cost is $0.001 to $0.01 per document using gpt-4o-mini, which is less than 5% of the savings from manual processing. Comparison of popular models:

from pydantic import BaseModel
from openai import OpenAI

class InvoiceData(BaseModel):
    vendor_name: str
    invoice_number: str
    total_amount: float
    currency: str
    due_date: str | None

client = OpenAI()
response = client.beta.chat.completions.parse(
    model="gpt-4o-mini",
    messages=[{"role": "user", "content": f"Extract invoice data:\n{text}"}],
    response_format=InvoiceData,
)

Metrics and monitoring

After prod launch, track:

• Extraction accuracy — percentage of correctly extracted fields (reference sample)
• Human escalation rate — target: reduce from 30–40% (manual) to 5–10%
• Processing latency — p95 LLM call time, target < 3 sec for sync processes
• Token cost per document — for budgeting, typically $0.001–0.01 per document with gpt-4o-mini

Typical results after deployment: document processing time drops from 3–5 minutes (manual) to 15–30 seconds, accuracy on structured fields reaches 92–96%. Our experience: over 10 years in AI/ML, completed 50+ RPA and LLM integration projects. Our company has been in the AI automation market for over 5 years. We'll evaluate your project in one day — contact us for a consultation. Get advice on architecture and model selection.

Implementation timelines

Steps to implement:

1. Collect sample documents (10-20 per type).
2. Configure LLM extraction with few-shot prompting.
3. Deploy microservice and integrate with RPA.
4. Monitor accuracy and adjust prompts.
5. Scale to full production with fallback.

• Prototype (1 document type, 1 process): 2–3 weeks
• MVP (3–5 document types, CRM/ERP integration): 6–8 weeks
• Scalable solution (queue, monitoring, fallback): 10–14 weeks

Our project data (2024) shows LLM integration outperforms traditional regex parsing by 3x in accuracy.

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.