AI Dispatcher – Digital Employee for Task Coordination

AI Dispatcher – Digital Employee for Task Coordination

A dispatch service drowning in tickets: operators manually combing through emails and chats, assigning engineers on a hunch, SLA breaches piling up. We've seen this firsthand in telecom, handling 800 daily field service requests. Manual dispatching couldn't keep up: average task assignment time was 12 minutes, individual engineer overload reached 30%, and SLA compliance lingered at 71%. After deploying an AI dispatcher, assignment time dropped to 45 seconds, SLA compliance rose to 89%, and engineer load evened out. The classification model fine-tunes on your historical data, improving accuracy each month.

Our digital dispatcher automatically receives incoming requests, classifies them, assigns assignees based on workload and competencies, tracks execution, and escalates overdue tasks. All without human intervention. We implement such systems turnkey—from classifier configuration to CRM integration.

How does the AI dispatcher route requests?

At the core is intelligent intake and routing. The process consists of three steps: request classification, assignee scoring, and execution monitoring. Let's walk through the code:

from openai import AsyncOpenAI
from pydantic import BaseModel
from typing import Literal, Optional

client = AsyncOpenAI()

class TaskClassification(BaseModel):
    category: str
    priority: Literal["critical", "high", "normal", "low"]
    required_skill: str
    estimated_duration_minutes: int
    location: Optional[str]
    sla_hours: float
    special_requirements: list[str]

class DispatcherAgent:

    def __init__(self, team_db, task_db):
        self.team_db = team_db
        self.task_db = task_db

    async def process_incoming_request(self, request: dict) -> dict:
        cls = await client.beta.chat.completions.parse(
            model="gpt-4o",
            messages=[{
                "role": "system",
                "content": "Classify the incoming request for dispatching. Determine priority, category, required competencies, and expected duration."
            }, {
                "role": "user",
                "content": f"Request: {request['description']}\nFrom: {request['client']}\nContact: {request['contact']}"
            }],
            response_format=TaskClassification,
            temperature=0,
        )
        task_class = cls.choices[0].message.parsed
        assignee = await self.select_best_assignee(task_class, request.get("location"))
        task = await self.task_db.create({
            "title": request.get("title", f"Request from {request['client']}"),
            "description": request["description"],
            "category": task_class.category,
            "priority": task_class.priority,
            "required_skill": task_class.required_skill,
            "assignee_id": assignee["id"] if assignee else None,
            "sla_deadline": datetime.now() + timedelta(hours=task_class.sla_hours),
            "status": "assigned" if assignee else "pending",
        })
        if assignee:
            await self.notify_assignee(assignee, task)
        return {"task_id": task["id"], "assignee": assignee, "sla": task["sla_deadline"]}

    async def select_best_assignee(self, task, location):
        available = await self.team_db.get_available(skill=task.required_skill, shift="current")
        if not available:
            return None
        scored = []
        for person in available:
            score = 100
            current_load = await self.task_db.count_active(person["id"])
            score -= current_load * 10
            if location and person.get("current_location"):
                distance = calculate_distance(location, person["current_location"])
                score -= min(distance / 10, 30)
            if task.required_skill in person.get("specializations", []):
                score += 20
            scored.append({**person, "score": score})
        return max(scored, key=lambda x: x["score"]) if scored else None

In the first step, the GPT-4o model analyzes the request description and extracts category, priority, and required skills. The scoring algorithm uses a base of 100 points: penalties for workload (10 points per active task) and distance (up to 30 points), bonus for specialization (20 points). This ensures optimal distribution given the current situation.

Why is SLA monitoring critical?

Every task has a Service Level Agreement (SLA)—a completion deadline. Breaching it damages business reputation. Our AI dispatcher doesn't just create tasks; it monitors them in real time. Every 15 minutes, the system checks remaining time until deadline: if less than an hour, a reminder is sent; under 30 minutes, an urgent notification; a breach triggers escalation to management. Monitoring code:

class SLAMonitor:

    async def check_and_escalate(self):
        tasks = await self.task_db.get_active_tasks()
        now = datetime.now()
        for task in tasks:
            sla_deadline = task["sla_deadline"]
            time_to_sla = (sla_deadline - now).total_seconds() / 3600
            if time_to_sla < 0:
                await self.handle_sla_breach(task)
            elif time_to_sla < 0.5:
                await self.send_urgent_reminder(task)
            elif time_to_sla < 1 and task["status"] == "assigned":
                await self.escalate_to_supervisor(task, reason="task_not_started")

    async def handle_sla_breach(self, task):
        breach_message = await client.chat.completions.create(
            model="gpt-4o-mini",
            messages=[{
                "role": "system",
                "content": "Create a concise, concrete SLA breach escalation message for the manager."
            }, {
                "role": "user",
                "content": f"Task: {task['title']}, SLA breached by {abs(int((datetime.now() - task['sla_deadline']).total_seconds() / 60))} min"
            }],
        )
        await slack_client.post_message(
            channel="#dispatcher-escalations",
            text=f"🚨 SLA breached:\n{breach_message.choices[0].message.content}",
        )
        await self.task_db.update(task["id"], {"sla_status": "breached"})

How to train the request classifier?

To train the classifier, we use your historical data. Steps:

1. Dataset collection: at least 500 labeled requests.
2. Preprocessing: cleaning, lemmatization.
3. Model tuning: fine-tuning GPT-4o or other LLM on classification tasks.
4. Validation: check accuracy on a held-out set (minimum 85%).
5. Deployment: upload the model to SageMaker or Vertex AI.

What results did the AI dispatcher show in telecom?

From our practice: a telecom operator with field engineers, 800 field service requests daily. Before: manual distribution, 12 minutes per assignment, 71% SLA compliance. After: AI dispatcher.

The AI dispatcher assigns tasks 16 times faster than a human, and SLA compliance improves by 18 percentage points. The system accepts requests via chat, email, CRM API; classifies them into 8 types (connection, repair, equipment replacement); assigns an engineer based on skills, load, geolocation; monitors SLA (4 hours for critical, 24 hours for standard); automatically reminds and escalates overdue tasks.

What's included in AI dispatcher development

• Request classifier: configuring categories, priorities, required skills
• Assignment algorithm: scoring assignees by workload, skills, geolocation
• SLA monitoring: automatic checks, reminders, escalations
• Integration: with CRM, messengers (Slack/Telegram), notification APIs
• Documentation: architecture description, admin guide
• Training: classification model fine-tuning on your data
• Support: 3 months warranty after launch

Implementation timelines

Contact us to discuss implementing an AI dispatcher in your service. Get a consultation on dispatching optimization. Our experience: certified engineers, work with major telecom operators.

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.