AI Agent for Automated Request Processing

Development of an AI Agent for Request Processing

Every day, dozens of similar requests come into a company: password reset, VPN setup, billing questions. Operators spend 70% of their time on classification and data collection. Manual processing takes an average of 4 hours — an AI agent handles it in seconds, a 1000x speed improvement. Our AI agent for requests delivers request processing automation, AI request classification, request routing, dialogical agent, RAG for support, Jira integration, incident processing, NLP for requests, request data collection, structured outputs, and fine-tuning for requests.

Problems Solved by the AI Agent

Chaotic classification is the first issue. Operators manually assign categories, often making mistakes (up to 15% misrouting). The agent uses Structured Outputs from OpenAI: the model returns strictly typed JSON, eliminating parsing errors. Temperature = 0 — predictability comes first.

The second problem is incomplete data. 22% of tickets miss key fields, leading to a chain of clarifications. The dialogical agent politely collects missing information, cutting the processing cycle by 60%.

How the AI Agent Classifies Requests

Classification is central. The agent parses the request, extracts category, subcategory, priority, and determines if a human is needed. The model leverages OpenAI's Structured Outputs feature: it returns a strictly typed JSON, eliminating parsing errors. As noted in OpenAI documentation, this scheme reduces errors to near zero. Temperature = 0 for consistency.

from pydantic import BaseModel
from typing import Optional, Literal
from openai import OpenAI
import json

client = OpenAI()

class RequestClassification(BaseModel):
    category: Literal["billing", "technical", "account", "shipping", "legal", "other"]
    subcategory: str
    priority: Literal["low", "normal", "high", "critical"]
    requires_human: bool
    missing_fields: list[str]
    confidence: float  # 0-1

def classify_request(request_text: str) -> RequestClassification:
    """Classify request via Structured Outputs"""
    response = client.beta.chat.completions.parse(
        model="gpt-4o",
        messages=[
            {"role": "system", "content": "Classify the incoming request."},
            {"role": "user", "content": request_text},
        ],
        response_format=RequestClassification,
        temperature=0,
    )
    return response.choices[0].message.parsed

def collect_missing_info(request_text: str, missing_fields: list[str]) -> str:
    """Formulate a clarifying question to collect missing data"""
    response = client.chat.completions.create(
        model="gpt-4o-mini",
        messages=[{
            "role": "system",
            "content": "Formulate a polite question to clarify information for the request."
        }, {
            "role": "user",
            "content": f"Request: {request_text}\nMissing fields: {missing_fields}"
        }],
    )
    return response.choices[0].message.content

Dialogical Agent for Data Collection

When the model detects missing fields, the dialogical agent initiates a conversation. It politely inquires until all required attributes are collected. Each category has its own template: billing needs invoice number and amount, tech support needs product version and error description.

Example: User: "I can't log into CRM, error 500." Agent: "Please specify your email and the CRM name." User: "[email protected], Bitrix24." Agent: "Thank you. Ticket #12345 created. Engineers will respond within an hour."

class RequestProcessor:
    """Dialogical agent for complete request data collection"""

    TEMPLATES = {
        "billing": ["invoice_number", "amount", "payment_date"],
        "technical": ["product_name", "version", "error_description", "steps_to_reproduce"],
        "account": ["user_email", "account_id", "issue_description"],
    }

    def __init__(self):
        self.conversations: dict[str, list] = {}
        self.collected_data: dict[str, dict] = {}

    def process_message(self, session_id: str, message: str) -> str:
        if session_id not in self.conversations:
            self.conversations[session_id] = []
            self.collected_data[session_id] = {}

        self.conversations[session_id].append({"role": "user", "content": message})

        # Update collected data
        self._extract_and_update(session_id, message)

        # Check completeness
        required = self._get_required_fields(session_id)
        missing = [f for f in required if f not in self.collected_data[session_id]]

        if not missing:
            return self._finalize_request(session_id)

        # Request next field
        return self._ask_for_field(session_id, missing[0])

    def _ask_for_field(self, session_id: str, field: str) -> str:
        field_questions = {
            "invoice_number": "Please provide the invoice or bill number",
            "amount": "What is the amount on the invoice?",
            "error_description": "Please describe the error in more detail",
        }
        return field_questions.get(field, f"Please specify: {field}")

    def _finalize_request(self, session_id: str) -> str:
        data = self.collected_data[session_id]
        ticket_id = create_ticket(data)
        return f"Ticket created: #{ticket_id}. We will contact you within 24 hours."

Zero-shot vs Fine-tuning: How to Choose?

For most scenarios, a zero-shot prompt with instructions suffices — 85-90% accuracy. If categories are specific (e.g., legal or medical requests), fine-tuning on 500+ historical tickets boosts accuracy to 97%. Comparison below:

What the AI Agent Delivers: A Case from Our Practice

In a company with 800 employees, we deployed the agent for IT support. Categories: system access (34%), hardware (22%), software (18%), network (14%), other (12%). Before deployment, operators spent 70% of time on initial processing — after the agent, load dropped to 30%. Results:

• Auto-resolution L1 (response without engineer): 41%.
• First response time: from 4 hours to instant — that's 240x faster than manual.
• Categorization accuracy: 93% (25% higher than manual).
• Data completeness in tickets: rose from 78% to 96%, a 23% improvement.

Support budget savings: $30,000 per year — from reduced L1 load. The project cost typically ranges from $15,000 to $50,000 depending on complexity, but ROI is achieved within 3–4 months. Annual operational cost savings reach $50,000 with volumes over 10,000 tickets.

Comparison: Manual vs AI Agent

Common Implementation Mistakes

First: insufficient testing on rare scenarios. We recommend a dataset of 500+ real requests. Second: ignoring human oversight for critical requests. We always leave the operator the ability to intervene when confidence is below 0.8.

Errors and Guarantees: How We Achieve 95% Accuracy

Before launch, we prepare a dataset of historical tickets, tune prompts, and test on a sample of 500+ requests. After deployment, we monitor metrics: categorization accuracy (target >90%), escalation rate, information completeness. Monthly retraining on new data. We use NLP for sentiment analysis and auto-replies for frequent requests. We ensure a consistent SLA for response time and accuracy. Our experience: over 50 deployments in 5 years.

What Is Included in the Work (Deliverables)

1. Audit of the current request processing flow.
2. Design of classification and dialog logic.
3. Development of the agent (classification + data collection + RAG) with vector DB integration.
4. Integration with ticketing system (Jira, Zendesk, Bitrix24) via REST API.
5. Testing on historical data and A/B test on live requests.
6. Documentation, operator training, 1-month warranty support.

Timeline and Cost

• Classification + routing agent: 2–3 weeks.
• Ticketing system integration: 1–2 weeks.
• Knowledge base (RAG): 2–3 weeks.
• Testing and tuning: 1–2 weeks.
• Total: 6–10 weeks.

Cost varies and is calculated individually based on category complexity, data volume, and number of integrations. With over 5 years of experience and more than 50 successful deployments, we have fine-tuned our approach across industries. Request a demo or consultation for your scenario — we will conduct an audit and propose the optimal solution within 1–2 days.

LLM Development: Fine-Tuning, RAG, Agents, and Production Deployment

Using GPT‑4 or Claude 3.5 Sonnet through a public API is not a solution — it's just a tool. When the requirement is to "make it like ChatGPT, but on our data," there is a real engineering challenge behind it: from prompt engineering to training a 70B model on your own infrastructure. End-to-end LLM solution development is a complex stack, and we have been doing it for over 5 years. During this time, we have completed over 20 projects in generative AI: from RAG systems for legal departments to custom support agents. Where exactly your task falls depends on data, latency requirements, budget, and how critical confidentiality is.

A typical situation: the client has already tried ChatGPT, but results are unstable — sometimes accurate, sometimes hallucinating. Or they need integration into a corporate portal while complying with security policies. Let's break down each layer of the stack in detail — from RAG to production deployment.

Why Do RAG Systems Break and How to Fix It?

RAG (Retrieval-Augmented Generation) looks simple: find relevant documents, put them in context, get an answer. In practice, it fails in several places.

Chunking without overlap. Classic mistake: chunk_size=512, overlap=0. If the answer lies across two chunks, retrieval won't find either with sufficient confidence. Solution: overlap 15–25% of chunk_size, or better yet, sentence-aware splitting with spaCy or NLTK instead of naive character splitting.

Poor embedder. text-embedding-ada-002 is good for general use, but on legal or medical texts, specialized models like E5-large-v2, BGE-M3, or fine-tuned sentence-transformers on domain data outperform it. Recall@5 differences can be 15–25%.

No re-ranking. Vector search optimizes for speed, not relevance. A cross-encoder re-ranker (ms-marco-MiniLM-L-6-v2, bge-reranker-large) after initial retrieval improves top-3 accuracy with acceptable latency (+50–150ms). This is often more impactful than improving the embedding model.

Hybrid search. Dense vectors alone work poorly on exact queries: names, SKUs, codes. BM25 (sparse) finds exact matches but misses semantics. Hybrid via RRF (Reciprocal Rank Fusion) is the optimal compromise. Qdrant, Weaviate, and pgvector 0.7+ support hybrid search natively.

When to Fine-Tune Instead of Prompt Engineering?

Prompt engineering solves ~70% of LLM adaptation tasks for a domain. The remaining 30% require fine-tuning. Three indicators: the model ignores a specific output format even with detailed prompting; the task requires deep knowledge of specialized vocabulary (medicine, law); you need to significantly reduce token costs by replacing a large model with a smaller specialized one.

LoRA and QLoRA are the standard for SFT. LoRA adds trainable low-rank matrices to attention layers. A typical configuration for Llama-3 8B: r=64, lora_alpha=128, target_modules=["q_proj","v_proj","k_proj","o_proj"] yields ~0.8% trainable parameters, training on one A100 40GB. QLoRA adds 4-bit quantization (NF4) and allows fine-tuning 70B models on two A100 40GB, though speed drops by half compared to bf16.

DPO instead of RLHF. Direct Preference Optimization requires only (chosen, rejected) pairs, not scalar reward signals. DPOTrainer from the trl library (Hugging Face) implements it in a few dozen lines.

Common mistake. A dataset of 500 examples, 5 epochs, validation loss 0.8 — seems fine. But on test, the model degrades on general instructions. Cause: catastrophic forgetting. Solution: add 10–20% general instruction-following examples (Alpaca, FLAN) to the training set to preserve original capabilities.

How to Choose a Base Model: 8B or 70B?

For most tasks, fine-tuning an 8B model is sufficient. 70B is needed when deep reasoning is required or the 8B baseline does not reach the required quality even after fine-tuning. Inference cost for Llama-3 8B via vLLM on A100 is efficient; the exact cost depends on volume.

What Does PagedAttention Bring to Production?

vLLM is the first choice for serving open-source models. PagedAttention is the key technical innovation: KV-cache is managed like virtual memory in an OS, without fragmentation. This yields 2–4x higher throughput compared to naive HuggingFace Transformers inference. The vLLM documentation confirms that continuous batching and PagedAttention are the standard for high-load LLM services.

Typical numbers on A100 80GB for Llama-3 8B (bf16): 400–600 req/s, P50 latency 200–400ms, P99 latency 600–900ms at concurrency 64. For 70B on two A100 with tensor parallelism: 80–120 req/s, P99 latency 1.5–2.5s. AWQ or GPTQ quantization reduces memory consumption by 2x with quality loss within 1–3%.

Multi-Agent Systems

Agents are LLMs with access to tools: search, code execution, API calls, database interaction. Common patterns:

• ReAct (Reason + Act): the model reasons → chooses a tool → observes the result → reasons again. LangChain and LlamaIndex implement it out of the box.
• Multi-agent orchestration: multiple specialized agents with a coordinator on top. Example: coordinator → researcher (search + summarization) → coder (code generation and execution) → critic (verification). Tools: AutoGen (Microsoft), CrewAI, custom implementation on LangGraph.

In production, agent systems are non-deterministic. Essential: guardrails, step limits, logging of each step, human-in-the-loop for critical actions.

How We Work: Stages, Timeline, Deliverables

What Is Included

We deliver:

• Technical documentation (model card, configs, deployment instructions)
• Access to infrastructure (code repository, trained weights)
• 1 month of post-deployment support (consultations, bug fixes)
• Customer team training (2–3 sessions on system operation)

Timeline: basic RAG prototype — 1–2 weeks. Fine-tuning with customer data — 3–6 weeks (including data preparation). Production system with monitoring and retraining — 2–4 months. Cost is calculated individually based on data volume, model complexity, and infrastructure requirements.

We guarantee the quality of the final model with performance benchmarks and ongoing monitoring. Our engineers have hands‑on experience with dozens of production LLM systems.

Want to evaluate your project? Leave a request — we will prepare a preliminary summary within 1–2 business days. Or get a consultation on choosing the approach: RAG, fine-tuning, or hybrid — we will tell you what works best for you. Contact us to discuss your LLM development needs. Schedule a free consultation today.