What is an LXP? A complete introduction to Karzaan’s smart learning experience platform

For decades, corporate training was a one-way process: the organization decided and the employee complied. One fixed path, the same content, one standard exam, for everyone. The result? Low completion rates, shallow learning, and systems that exist more for management reporting than for real skill development. Karzaan’s smart learning experience platform (LXP) changes this equation at its root. Instead of one path for everyone, a smart coach with 11 specialized AI agents stays with the user from the moment they sign in until they receive their certificate: it asks, diagnoses, teaches, assesses, adapts, and suggests a review at the very moment forgetting begins. In this article, we present the complete architecture of Karzaan’s LXP in full detail.

First things first: what is the difference between an LXP and an LMS?

As we covered in detail in our article on the difference between LMS and LXP , and explained in our article on combining LMS and LXP why the two belong together, here is a quick recap:

Controlling and managing training
Approach
A personalized experience
Predefined, the same for everyone
Learning path
Dynamic, unique to each person
Limited or passive
Role of AI
Central at every stage
Training manager
Focus
The learner
End-of-course exam
Assessment
Continuous, adaptive assessment
None
Managing forgetting
Automatic spaced repetition
Usually none
Practical simulation
Scenarios and role-play
Usually standard
Data security
Dedicated organizational filter

Karzaan’s LXP is designed not as a replacement for an LMS but as an intelligent layer on top of it. This means mandatory and compliance training is still controlled and reported through the LMS infrastructure, while each person’s learning experience is personalized through the LXP.

Why is Karzaan’s LXP different from other platforms?

Many platforms that claim to be an LXP today are really an LMS with a simple content recommendation engine. Karzaan’s key difference is that the system is made up of 11 specialized AI agents, each responsible for one part of the learning process, together forming a coordinated ecosystem rather than a single algorithm trying to do everything.

These 11 agents are organized into 4 main phases: 

LXP and its agents

Phase 1: Discovering and understanding the user

Every successful learning journey starts with an accurate understanding. Before showing the user even the smallest piece of content, the first phase of Karzaan’s LXP does three essential things: it asks, it assesses, and it recommends.

1

Agent 1: Dynamic Interviewer

Input

Basic user profile (education, experience, current role) The user’s conversational answers
1 Dynamic Interviewer

Output

Map of learning goals and interests Strengths and weaknesses as seen by the user Initial structure for personalization parameters

Design philosophy

A standard sign-up form only collects structured data. But what really determines the learning path, motivation, fears, real goals, hidden strengths, comes out in a natural conversation.

How it works

Instead of rigid forms, this agent holds a smart conversation with the user. It asks about career goals, listens to the answers, designs the next question based on them, and extracts hidden patterns such as learning anxiety, genuine interests or skill gaps from the responses.

User experience

It feels like sitting with a professional advisor, not filling in a form.

2

Agent 2: Pre-assessment Diagnoser

Input

User’s job level Standard question bank Agent 1’s answers
2 Pre-assessment Diagnoser

Output

Objective baseline score Map of “blind spots” (topics where the user is weak) Recommended content difficulty level

Design philosophy

Nothing kills motivation like content that is too easy or too hard. For the system to know where to start, it first has to know where the learner is now.

How it works

It runs a standard diagnostic test (usually 10 questions) that requires no prior study. The goal is to measure the user’s current knowledge, not to test short-term memory. Questions are adaptive: if the user answers the first question correctly, the second one gets harder, and vice versa.

User experience

The user sees content that is neither boringly familiar nor confusingly indigestible, but right in their “growth zone”.

3

Agent 3: Resource Advisor

Input

Results from agents 1 and 2 Metadata and summaries of all organizational resources
3 Resource Advisor

Output

3 prioritized recommended resources A clear reason for each recommendation Map of the skills each resource strengthens

Design philosophy

Large organizations can have hundreds of learning resources. Choosing among them without guidance often leads to picking the most familiar or easiest resource, not the most useful one.

How it works

Using the results of the two previous agents, it recommends the top three resources from all of the organization’s available materials (PDFs, videos, podcasts, online courses). More important than the recommendation is the explanation: “I recommend this resource because it strengthens skill X, which was weak in your test, and matches career goal Y that you mentioned in the interview.”

User experience

The user makes an informed choice and knows why this resource was selected for them.

Phase 2: Planning and teaching

Once the user is understood and a resource is chosen, the second phase starts the core work of learning.

4

Agent 4: Smart Planner

Input

Full text of the resource with its page structure User’s time profile (how many hours a day?) User level (from agent 2)
4 Smart Planner

Output

Step-by-step study plan Outline of each session + page range Specific learning goals for each session

Design philosophy

A 200-page learning resource dumped on the user all at once has no learning effect. Breaking it down into digestible units is this agent’s main job.

How it works

It analyzes the chosen resource and splits it into specific sessions based on two main parameters: the user’s level and their available daily time. Not an arbitrary split by page count, but a logical breakdown by topic, conceptual unity and the right cognitive load for each session.

User experience

The user knows what they will study today, what comes tomorrow, and where the finish line is.

Practical example

A 150-page PDF for a user with 45 minutes a day is split into 10 sessions: “Session 1: pages 1–18, topic: basics and core terminology, goal: learn 15 key concepts”.

5

Agent 5: Interactive Teacher

Input

Content specified by the planner User level and profile User questions and requests
5 Interactive Teacher

Output

Structured content with a title, summary, text and key points Interactive answers to questions Suggestion to continue or review

Design philosophy

This is the beating heart of the whole system. A textbook transfers information; a good teacher creates learning, and the difference is interaction.

How it works

It takes the raw content of each session and teaches it anew. That doesn’t mean restating the content, but rather:

  • Breaking it into subtopics: Each session is divided into 7–8 smaller blocks
  • Simplifying the language: Technical language becomes easy to understand without losing depth
  • Creating examples: Abstract concepts are explained with examples related to the user’s job role
  • Answering questions: The user can ask a question at any moment
  • Smart repetition: If the user says “I didn’t get it”, only that subtopic is repeated with a different approach

User experience

Dynamic teaching, like a private tutor who is patient, repeats as often as needed and explains the content in the user’s own language.

Important technical note

This agent’s output is produced for three formats at once: text (for reading), an audio summary (for podcasts) and flashcards (for quick review).

Phase 3: Monitoring, interaction and reinforcement

Real learning begins after studying. The third phase makes sure the content settles into long-term memory, not just short-term memory.

6

Agent 6: Real-time Quizzer

Input

Content of the same subtopic or session Desired number of questions User’s previous performance
6 Real-time Quizzer

Output

Questions with the right difficulty level Detailed answer key Weak-point analysis for the supervisor agent

Design philosophy

Learning research shows that active recall, trying to remember material instead of rereading it, makes learning 2 to 3 times more effective.

How it works

It generates quiz questions in two modes:

  • Mode 1, end of session: A formal quiz to measure mastery of the session content, combining multiple-choice, fill-in-the-blank and short-answer questions.
  • Mode 2, mid-session (random): One or two surprise questions during teaching that keep the learner focused and eliminate the “illusion of understanding”.

User experience

Active learning that locks content into long-term memory.

7

Agent 7: Role-Play / Scenario Builder

Input

Session topic and target skill User’s emotional intelligence level (from the profile)
7 Role-Play / Scenario Builder

Output

Realistic interactive scenario Qualitative assessment of the user’s responses Constructive feedback and improvement tips

Design philosophy

Theoretical knowledge doesn’t become a skill until it is applied in a real situation. For jobs that involve human interaction (sales, support, conflict management), this gap between knowledge and skill does the most damage.

How it works

Based on the session topic, it designs a realistic scenario and plays the other party (customer, colleague, manager). The user has to respond, and the system gives feedback.

User experience

Turning theoretical knowledge into practical skill in a safe environment, without the risk of making mistakes in front of a real customer.

Practical example

If the session topic is “handling customer complaints”, the system plays an unhappy customer: “I ordered three days ago and it still hasn’t arrived. This is completely unacceptable.” The user replies, and the system evaluates the quality of the response based on complaint-handling principles.

8

Agent 8: Spaced Repetition Scheduler

Input

Scores of all completed quizzes Date each topic was last reviewed The optimal time interval for that specific topic
8 Spaced Repetition Scheduler

Output

Suggestion to add 10–15 minutes of review to upcoming sessions The exact topic that needs to be reviewed

Design philosophy

People naturally forget. Ebbinghaus’s forgetting curve shows that without review, 70% of new material is lost within 24 hours. Spaced repetition is the most scientific technique for overcoming this.

How it works

It analyzes the scores of all previous quizzes and the date each topic was last reviewed. Based on this, it determines which older topics the user needs to review in upcoming sessions, and exactly when.

User experience

Over 80% long-term retention, with minimal time spent on review.

Practical example

“In session 5, spend the first 10 minutes reviewing the topic "Types of accounts" from session 2; it was last reviewed 6 days ago and its quiz score was 60%.”

9

Agent 9: Supervisor

Input

Complete performance record from the start to the current session Scores, study time, number of repeat requests User interaction pattern
9 Supervisor

Output

Plan change instructions (number of sessions, content volume) Changing the teaching agent’s style Alerting the training manager when needed

Design philosophy

This agent is the strategic brain of the whole system. No human can analyze all of a learner’s performance data fast enough and adapt the learning plan in real time. That is exactly what the supervisor agent does.

How it works

It monitors the outputs of all the other agents, and when it sees warning patterns, it adapts the plan:

  • Completion rate dropped? It reduces the session length
  • Quiz scores consistently high? It increases the difficulty (the user is ready for more challenge)
  • User keeps repeating a topic? It rewrites that topic with a completely different explanation
  • Study time decreased? It shifts the teaching style toward shorter, more engaging content

User experience

The system always keeps the user in their “growth zone”: not boringly easy, not frustratingly hard.

Phase 4: Final assessment and security

10

Agent 10: Summative Proctor

Input

All course content User’s performance record
10 Summative Proctor

Output

Comprehensive adaptive exam Final score + detailed analysis Certificate or remedial plan

Design philosophy

The end of a course should record a real point of progress, not just a certificate of attendance. The final assessment should be comprehensive (covering all sessions), fair (based on each user’s actual path) and constructive (even when the user doesn’t pass).

How it works

It designs a comprehensive exam covering all course topics. The difficulty matches the user’s actual level (not the same exam for everyone). Based on the result:

  • Score above the pass mark: A skills certificate is issued, along with an analysis of strengths
  • Score below the pass mark: A targeted remedial plan is suggested that focuses only on weak topics, rather than repeating the whole course

User experience

An official result that can be presented to managers, with genuinely useful feedback.

11

Agent 11: Security Guardrail / Processor

Input

Raw organizational file User question or request
11 Security Guardrail / Processor

Output

Standardized content, ready for teaching Filtered, secure, anonymized request Reports of suspicious activity to the system administrator

Design philosophy

This agent works behind the scenes; the user doesn’t interact with it directly, but without it no organization could use this system with peace of mind.

It works in three layers

  • Layer 1, content pre-processing: Organizational documents (PDF, Word, slides) are processed before entering the system. Tables, formulas, charts and images are extracted and converted into a format the AI model can understand.
  • Layer 2, content filter: User questions and requests are checked before being sent to the main model. Requests outside the learning scope (such as personal or unrelated questions) are detected and receive a predefined answer without being sent to the model.
  • Layer 3, data anonymization: The user’s identity information (name, employee ID, personal details) is replaced with unique identifiers before being sent to the AI web service. This means that even if data leaks in transit, it cannot be traced back to any real person.

Organization’s experience

Confidence that the organization’s confidential data is used in a controlled way and that employees’ privacy is protected at every stage.

Smart architecture Personalized learning system

Each user’s learning path is designed and run step by step, fully personalized, by a chain of AI agents; from sign-in to the final certificate, every decision is based on the user’s actual performance data.

Tap any agent to see its role and connections

1Discovery
2Planning
3The cycle of every session

For every learning session this cycle repeats in each session

4Final assessment
or

Personalized learning system

Tap any agent to see its role and connections

Which organizations is Karzaan’s LXP best suited for?

As we explained in our article on combining LMS and LXP , Karzaan’s LXP is designed to complement an existing LMS infrastructure. It creates the most value in these situations:

Organizations with a large volume of diverse content

Key agents

Industries where required skills change quickly (technology, banking, insurance, sales)

Key agents

Organizations facing low completion rates in their LMS

Key agents

Organizations that want to move from “mandatory training” to a “learning culture”

Key agents

Organizations whose employees need practical soft skills (not just theoretical knowledge)

Key agents

Integration with other systems

With HRM systems

Job role, career path and performance review data flow automatically into the learner profile, making the resource advisor agent’s recommendations more accurate.

With the corporate portal

Learning recommendations and progress are visible directly on the portal dashboard, with no need to sign in to a separate system.

LXP Karzaan smart LXP

With a mobile app

Employees can access courses through the mobile app while commuting or between meetings. The audio content (podcasts) is produced exactly for this scenario.

With the LMS infrastructure

Mandatory and compliance training is still controlled and reported through the LMS. The LXP sits on top of this infrastructure and adds the personalization layer.

Key performance indicators (KPIs) you can expect

Before LXP Target after implementation

Completion rate of optional courses

Before LXP20–35%
Target after implementation65–85%

Knowledge retention (after 30 days)

Before LXP25–35%
Target after implementation75–85%

Time to reach the desired skill level

Before LXP6–12 weeks
Target after implementation3–6 weeks

User satisfaction with the learning system

Before LXPMedium
Target after implementationHigh

Voluntary content usage rate

Before LXPLow
Target after implementationMedium to high

Conclusion

11 smart agents
4 coordinated phases
1 dedicated path for every learner

Karzaan is made up of 11 intelligent agents that, across 4 coordinated phases, turn learning from a uniform path into personalized coaching. From the first conversation to the final certificate, every step is designed based on that specific learner’s real data. The system is designed not to replace human teaching or a traditional LMS, but to close a gap that has existed in corporate training for years: the gap between “completing a course” and “actually learning”. It is worth noting that these agents form the core of an LXP (learning experience platform) and can be expanded or reduced according to each organization’s specific needs.

The gap Karzaan’s LXP closes

Completing a course Real learning

Want to see Karzaan’s LXP in your organization?

Request a consultation & demo

Frequently Asked Questions

01 Is the number of agents in Karzaan’s LXP fixed?

In Karzaan’s LXP, the core of 6 base agents (interviewer, assessor, advisor, planner, teacher and supervisor) is always present and active, since these are the pillars of any personalized learning system. The other 5 specialized agents (quizzer, simulator, spaced repetition, proctor and security filter) can be removed, disabled or replaced with more advanced agents as the organization needs. This flexibility makes Karzaan’s LXP an ideal choice for organizations of any size, industry and level of digital maturity. You are never locked into a fixed set of agents and can always align the system with your current needs.

02 Does the organization’s learning content need a specific format?

The security processor agent (no. 11) handles a wide range of formats (PDF, Word, PowerPoint). The content can be what the organization has already produced.

03 Where is user data stored?

User profiles and learning results are stored on the organization’s own servers (on-premise) or in a private cloud. Before any data is sent to the AI web service, the security agent removes identity information.

04 Is this system suitable for small organizations too?

The full value of Karzaan’s LXP shows when there is enough learning content and enough learners. For smaller organizations, we recommend starting with a custom LMS and adding the LXP layer at a later stage.

05 Can the agents be customized to the organization’s needs?

Yes. The agent-based architecture lets each agent be configured to the organization’s specific needs, from the type of role-play scenarios (agent 7) to the pass mark (agent 10) and content filter rules (agent 11).

06 Doesn’t using an LXP make employees feel monitored and controlled?

This is one of the most sensitive aspects of implementation. Karzaan’s LXP is designed so that learning data is used only to improve the user’s own experience, not for job performance evaluation or HR decisions. Being transparent with users about this boundary is one of the prerequisites for a successful implementation.

07 How long does it take to launch Karzaan’s LXP in an organization?

It depends on the complexity of integrating with existing systems and the volume of learning content. Our priority is usually to launch an initial pilot phase (with a specific department or group) as quickly as possible, so the organization can see real results before a full rollout.