ИИ-агент консультанта 1С

In this video, we explore a real-life case study of creating an AI agent for a 1C:UNF consultant—from the initial idea and prototype to industrial use by consultants at Aiton. We explain why businesses need such an agent, what problems it solves in customer support, how it helps reduce consultant workloads, speed up responses to common questions, and leverage the company's accumulated knowledge base more effectively. The presentation details: — Why a classic bot is insufficient for the complex subject area of ​​1C; — How the first prototype was created using the RAG approach; — Why datasets, quality metrics, and monitoring are needed; — Why an AI agent should be able to clarify questions, work with context, images, and external sources; — How an agent helps consultants respond to clients faster and train new specialists; — Why it's not a replacement for an expert, but a tool for strengthening the team. This is a practical example of how artificial intelligence can be used not just for hype, but to truly improve the efficiency of consulting, support, and management processes within a company. Timecodes 00:00 — Introduction and Topic 00:33 — Speaker's Experience: From Development to an AI-Driven Approach 02:14 — Why the Project Is Interesting for Business and Development 03:01 — Who the Client Is: Aiton and 1C:UNF Expertise 03:54 — Two Key Problems: Support Load and Consultant Training 06:34 — Solution Idea: An AI Bot That Relieves Consultants 07:29 — First Stage: Prototype and Hypothesis Testing 08:41 — Why Analytics, Metrics, and Monitoring Were Built In Right Away 09:26 — Data Analysis: Real Dialogues and a Knowledge Base 10:51 — Synthetic Dataset for Quality Assessment 12:21 — RAG Architecture: How a Question-Answering System Works 12:47 — Two Typical Mistakes in Creating AI Systems 14:01 — Simple Architecture as the Foundation for Managed Development 15:04 — User Feedback and Response Evaluation 15:26 — Monitoring System and Quality Measurement 17:01 — Prototype Results and RAGAS Metrics 18:15 — Agent Testing for New Consultants 19:10 — What Does a Bot Work in Telegram Look Like? 19:36 — Response Confidence, Sources, and Result Verifiability 20:46 — Why a Prototype Isn't Enough for Industrial Use 21:36 — Domain Complexity and Ambiguous Terms 22:38 — How to Answer Questions About Missing Features 23:21 — Using Videos and Additional Materials as Knowledge Sources 24:03 — Why an Agent-Based Architecture Was Needed 25:35 — An AI Agent as an Enhancement of the Classic RAG Approach 26:13 — What an Agent Consists of: Model, Context, Memory, and Tools 27:00 — Context Engineering and Working with Dialogue History 28:03 — The Role of System Prompts in Behavior Agent 28:49 — Clarifying Questions and Working with Terminology 29:34 — Security and Subject Area Limitations 30:04 — Agent Tools: Searching the Knowledge Base 31:01 — Working with Images and Screenshots 31:45 — Searching for Information in External Sources 33:00 — How the Quality of Agent Architecture Was Assessed 34:12 — The Role of Customer Experts in Preparing Datasets 35:06 — New Metrics for Assessing Agent Autonomous Performance 36:19 — Examples of the Agent as a True Consultant 36:28 — Clarifying the Meaning of Ambiguous Terms 36:53 — Working with 1C Interface Screenshots 37:52 — Answers to Missing Functions and Workarounds 38:51 — An Industrial Solution Based on AI-Driven Methodology 39:54 — How the Agent is Already Helping Aiton Consultants 40:11 — Speeding Up Client Responses and Training New Consultants Specialists 41:02 — When will the agent be available to clients? 41:29 — Conclusion of the presentation and contact information 42:17 — Commentary from "ITone" and questions from participants 42:47 — The idea of ​​an AI assistant within 1C:UNF 43:37 — Conclusion of the question block *********************************************************** Subscribe: Zen — https://zen.yandex.ru/itone Telegram channel: — https://t.me/itone_unf For questions, please contact: Our website — https://unf.itone.ru Telegram — https://t.me/ITone_Support_Bot WhatsApp — https://wa.me/74959891186 VKontakte — https://vk.com/itone_ru