AI glossary
The terms that come up in every automation meeting, explained in plain language with their practical meaning for a business.
AI agent
An AI agent is a program built on a language model that completes a task end to end: it understands the request, calls the systems it needs and returns a result. Unlike a scripted chatbot, an agent decides the sequence of steps itself instead of waiting for a button press. In Darobots projects, AI agents sell on Telegram, qualify leads and create orders in the CRM.
Learn moreChatbot
A chatbot is a program that talks to customers in text inside a messenger or on a website. A simple bot follows a button script, an AI bot understands free-form text. Bot development at Darobots starts at $200, and a basic consultant bot goes live in 2-3 days.
Learn moreLLM (large language model)
An LLM is a model trained on large volumes of text that predicts continuations and can therefore answer requests in natural language. GPT, Claude and Gemini are examples of LLMs. An LLM is what lets a bot answer a question that was never in the script.
RAG (retrieval-augmented generation)
RAG is an approach where the model searches your knowledge base for facts before answering and grounds its reply in what it found. It is the main way to make a bot talk about your real catalog, prices and documents instead of inventing them. Without RAG a model can only answer from what it saw during training.
Model hallucination
A hallucination is a confident but invented answer: a product that does not exist, a made-up price or condition. The fix is structure rather than a bigger model: connect the live catalog through RAG, restrict the topics and validate the answer before it reaches the customer.
Prompt and prompt engineering
A prompt is the instruction a model receives before answering: role, tone, topic boundaries and output format. Prompt engineering is the work of shaping that instruction so answers stay predictable. In business bots the prompt defines what the bot is allowed to promise a customer and what it is not.
Embeddings and vector databases
An embedding is a numeric representation of text that allows search by meaning rather than exact word match. A vector database stores those representations, so a bot can find the right product from a description like "something inexpensive for an office". This is the technical foundation of RAG.
Fine-tuning
Fine-tuning means further training an existing model on your own examples to lock in style and answer format. Most business tasks do not need it: connecting a knowledge base through RAG is cheaper, updates instantly and requires no retraining when the catalog changes.
Token
A token is the chunk of text a model operates on: roughly 3-4 characters. Both request length and model cost are measured in tokens, which is why the monthly cost of an AI bot depends on the number of conversations, not on the size of your customer base.
NLP (natural language processing)
NLP is the field that teaches software to work with human language: recognise intent, extract entities such as a date or address, classify requests. Inside a bot, NLP is what turns "tomorrow at 10 please" into an actual booking.
Telegram Mini App
A Telegram Mini App is a web app that opens inside Telegram with no installation. It is used where a chat is not enough: a catalog with photos, a booking date picker, a cart, a payment. The customer never leaves the messenger, so conversion is higher than sending them to a website.
Webhook
A webhook is how one system instantly notifies another about an event: a payment arrived, an order status changed, a lead came in. Most integrations between a bot, a CRM and a payment provider run on webhooks.
CRM integration
CRM integration is the link that moves data from a bot, website or form into a customer record automatically, with no manual re-entry. Darobots connects to an existing CRM (amoCRM, Bitrix24, HubSpot) or builds a custom one around the company's processes.
Learn moreBusiness process automation
Business process automation replaces manual steps with a chain that runs itself: a form submission lands in the CRM, the customer gets a confirmation and the manager gets a notification. Automating a single process at Darobots costs from $300 and goes live in 2-3 days.
Learn moreLead qualification
Lead qualification filters out requests that will never become deals before they reach a manager. An AI agent asks a few clarifying questions about the task, budget and timeline, and passes on only the people who are ready to buy.
AI audit
An AI audit is a walkthrough of a company's processes that shows where AI delivers measurable impact and where it is unnecessary. At Darobots it is a free 30-minute call that ends with an estimate of monthly savings before any work starts.
SLA
An SLA is a service level agreement: a fixed response time for incidents and a guaranteed availability level. For AI systems in production an SLA matters more than features: a bot that goes silent during peak orders costs more than a bot with fewer capabilities.
MVP
An MVP is the first working version of a system with the minimum set of features needed to test a hypothesis on real customers. Darobots ships a chatbot MVP in 2-7 days: testing demand on a working bot is cheaper than spending months agreeing on a full system.
Let's work out which of these technologies your business actually needs