Our AI Agents

Implementation period

---

2nd quarter of 2025.

Agent description

---

This is an intelligent admissions assistant for the Kazakhstan-American Free University (KAFU). It operates using a local LLM and RAG system to search for relevant answers to user queries.

Agent operation

---

Below are the main technologies and components that ensure the operation of the intelligent agent.

• 1️⃣ PostgreSQL database – used to store embeddings (vectors) of answers divided into text chunks;
• 2️⃣ Local model Qwen3-Embedding-0.6B – a multilingual model used to generate vector representations (embeddings) of texts;
  • 💬 Model context: 32,768 tokens.
• 3️⃣ Local model Google/Gemma-3-27B – a multilingual language model with tool_choice support, allowing integration of external tools such as web search and the ability to process data from the RAG system;
  • 💬 Model context: 7,500 tokens.
• 4️⃣ RAG and RAG-Fusion mechanisms – used to combine and rank results from multiple queries to form the most relevant context before generating a model response;
• 5️⃣ Reranking mechanism using the bge-reranker-v2-m3 model – used to select the most accurate and relevant answer from multiple retrieved fragments;
• 6️⃣ Speech recognition + RVC models – based on the local Vosk model, enabling speech-to-text conversion (currently in Russian) and generating human-like voice responses through the RVC server;
• 7️⃣ Final result – delivery of the final message in the following format:
  {Context}: %Result%.
  {Question}: %Question%.
  {Note}: Provide a response to the applicant.

Agent functions

---

• ♿ Inclusive functions
• 🌎 Multilingual response generation
• 🌐 Ability to perform web searches
• ⭐ Short summaries of uploaded files
• 💬 Answers to admission-related questions
• 🔊 Speech recognition (currently in Russian)
• 🛠️ Response generation from the LLM using the RAG system

Additional information

---

The agent is available 24/7 on Telegram and assists applicants, parents, and school students in learning everything about the admission process to our university.

Implementation period

---

2nd quarter of 2025.

Agent description

---

This is a library assistant that helps users find books, recommend new releases, create and compare tables, and automatically format bibliographic references according to GOST standards.

API

---

Currently, a temporary solution from OpenAI is used in the form of API access to models. In the near future, the system will transition to a local LLM. Table population testing on the local LLM is currently in progress.

Agent operation

---

The following are the main technologies and components that ensure the operation of the intelligent agent.

• 1️⃣ MySQL Database – After integrating the automated library system KABIS, the agent gained the ability to execute queries to the book database in order to provide responses to user requests.
• 2️⃣ API integration with RMEb and IPRSmart – Executes queries to these digital library resources to search for and retrieve books.
• 3️⃣ gpt-4o-mini model – A multilingual model that uses tool_choice to perform database queries and deliver search results;
  • 💬 Model context: 65,536 tokens.
• 4️⃣ Final result – The model answers basic library-related questions. In cases where a user requests specific books, the model determines when to call functions, and after execution, provides the final result to the user.

Agent functions

---

• ♿ Inclusive functions
• ⚡ Book search and retrieval from RMEb
• 📖 Book search and retrieval from KABIS
• 📙 Book search and retrieval from IPRSmart
• 📊 Creation of comparative tables
• 🌎 Multilingual answer generation
• 🙋🏻‍♂️ Personalized responses for each user

Additional information

---

The library assistant is available 24/7 on the university website and assists students, faculty, and researchers in finding the resources they need.

Implementation period

---

4th quarter of 2025.

Agent description

---

An AI platform for the automated creation of educational courses and recording lectures with AI avatars.

Agent operation

---

Below are the main technologies and components that ensure the operation of the intelligent agent.

• 1️⃣ Recording videos with AI teacher avatars — an integrated system for creating video lectures with realistic AI avatars. The avatars are generated based on photos and voice recordings of real instructors, creating a “live presence” effect and increasing student engagement and trust.
• 2️⃣ Generation of educational materials — the agent automatically generates lecture texts, presentations, methodological guides, and practical assignments for a given discipline. It uses university-standard templates and subject databases.
• 3️⃣ Realistic AI avatars — models for face generation and speech synchronization with a high degree of realism (up to 95%) are applied. This allows students to perceive the content as a live lecture, enhancing engagement and trust in the instructor.
• 4️⃣ Final result — the agent produces a complete multimedia course, including video lectures, methodological materials, tests, and reference links. The instructor only needs to review and publish the course.

Agent functions

---

• ♿ Inclusive functions
• 🎓 Automation of online course creation
• 🎥 Recording videos with AI teacher avatars
• 📝 Generation of lecture texts and methodological guides
• 💻 Integration of created video materials into the Moodle platform
• 🤖 High avatar realism level to enhance student trust

Additional information

---

The HeyGen AI platform is available 24/7, allowing users to create high-quality videos with virtual avatars, speech synthesis, and multilingual support at any time.

Implementation period

---

2nd quarter of 2025.

Agent description

---

This is an intelligent assistant for students of the Kazakhstan-American Free University (KAFU). It operates using a local LLM and RAG system to search for relevant answers to user queries.

In the near future, the bot will combine 2 in 1. He will be able to answer both students' questions and act as a psychological assistant who will assess the degree of threat of a particular application for psychological assistance.

Agent operation

---

Below are the main technologies and components that ensure the operation of the intelligent agent.

• 1️⃣ PostgreSQL database – used to store embeddings (vectors) of responses divided into text chunks;
• 2️⃣ Local model Qwen3-Embedding-0.6B – a multilingual model used to generate vector representations (embeddings) of texts;
  • 💬 Model context: 32,768 tokens.
• 3️⃣ Local model Google/Gemma-3-27B – a multilingual language model with tool_choice support, allowing the integration of external tools such as web search and processing of data from the RAG system;
  • 💬 Model context: 7,500 tokens.
• 4️⃣ RAG and RAG-Fusion mechanism – used to merge and rank results from multiple queries, forming the most relevant context before generating the model's response;
• 5️⃣ Reranking mechanism using the bge-reranker-v2-m3 model – used to select the most accurate and relevant answer from multiple retrieved fragments;
• 6️⃣ Speech recognition + RVC models – operates on the local Vosk model, providing speech-to-text conversion (currently in Russian) and human-like voice response generation via the RVC server;
• 7️⃣ Final result – delivery of the final message in the following format:
  {Context}: %Result%.
  {Question}: %Question%.
  {Note}: Provide a response to the student.

Agent functions

---

• ♿ Inclusive functions
• 🤝 Psychological assistance
• 📝 Requesting a certificate from the university
• 💵 Checking debts and tuition payments
• 🔊 Speech recognition (currently in Russian)
• 🔒 Retrieving login and password for Moodle and Platonus systems
• 💬 Providing answers to questions related to the learning process
• 🛠️ Response generation from the LLM using the RAG system

Additional information

---

The agent is available 24/7 on Telegram and helps students quickly access the necessary information about the university.

Implementation period

---

4th quarter of 2025.

Agent description

---

A digital assistant designed to help international students of the Kazakhstan-American Free University (KAFU) with adaptation. It answers questions about accommodation, visa support, the learning process, and student life.

Agent operation

---

The following are the main technologies and components that ensure the operation of the intelligent agent:

• 1️⃣ FastAPI — a web framework for building APIs and processing user requests;
• 2️⃣ Redis — used for temporary storage of chat history;
• 3️⃣ Local model Qwen3-Embedding-0.6B — creates embeddings and helps the system understand the semantic meaning of queries;
  • 💬 Model context: 32,768 tokens.
• 4️⃣ Local model Google / Gemma-3-27B — the main language model for generating responses to users;
  • 💬 Model context: 7,500 tokens.
• 5️⃣ LangChain — a library that enables interaction between local models and dialogue logic;
• 6️⃣ Final result — a personalized response to international student queries:
  {Question}: What study programs are available?
  {Answer}: The following academic programs are available: …

Agent functions

---

• 🎓 Answers to study-related questions
• 📝 Storing chat history in Redis
• 🏠 Consultation on housing and dormitory matters
• 🌍 Assistance with adaptation of international students
• 💬 Multilingual communication in English, Russian, and Kazakh
• 🧠 Response generation via local model and LangChain

Additional information

---

The agent is available 24/7 and hosted on the KAFU AI-Sana platform. Its purpose is to simplify the adaptation process for international students and make communication with the university more convenient and understandable.