SkillPath

Inspiration

Learning today is not personalized.

Most people follow generic courses, fixed curriculums, or random tutorials online. These systems ignore what a learner already knows, how they actually learn, and where they struggle.

Two people can have the same goal but completely different backgrounds, yet most platforms push them through the same path.

This creates three major problems:

learners waste time on skills they already know
real skill gaps remain hidden
progress is measured by course completion instead of true mastery

This problem becomes even larger in organizations and training programs, where teams are often pushed through the same learning material regardless of individual skill levels.

Existing tools focus on delivering content, not optimizing the learning process itself. Even modern AI tools mostly act as chat assistants or simple course recommenders. They do not analyze a learner’s history, model their skill state, and adapt their learning path over time.

We wanted to treat learning as a dynamic intelligence problem.

Instead of assigning static content, SkillPath studies the learner (their skills, learning history, pace, and behavior) and determines the next best step in their learning journey.

What it does

SkillPath is an AI-powered learning intelligence platform that generates personalized learning roadmaps based on a learner’s real data.

Users upload information such as:

skills
past learning history
work or project experience
behavior signals
LinkedIn/profile data
goals and timeline

The system builds a structured LearnerProfile and analyzes it to guide the learning journey.

SkillPath then:

maps the learner’s goal to a skill dependency graph
identifies skill gaps and bottlenecks
recommends the next best learning step using machine learning
generates practice checkpoints to test real understanding
retrieves targeted resources instead of full courses
updates skill mastery dynamically as learners progress
visualizes growth through a live analytics dashboard

SkillPath also supports organization workspaces.

Admins can create teams, invite members, and track learning progress across the group. Each learner receives a personalized roadmap while the organization dashboard aggregates:

team skill development
progress toward shared goals
pacing and bottlenecks

This enables bootcamps, companies, and instructors to track real skill development instead of course completion.

Why this matters

Most education technology platforms focus on content delivery.

SkillPath focuses on learning optimization.

Instead of asking “what should we teach?”, our system asks:

“What is the single most impactful step this learner should take next?”

By modeling skills, mastery, and learning behavior, SkillPath can guide learners through efficient skill development rather than static courses.

This approach has applications in:

professional upskilling
corporate training
higher education
career transitions

How we built it

SkillPath runs an end-to-end learning intelligence pipeline.

1. Profile Ingestion

Learner data is uploaded and normalized into a structured LearnerProfile model using a backend built with Express + TypeScript + MongoDB.

Signals such as skill scores, learning history, and behavior patterns are extracted and standardized for downstream analysis.

2. Goal Interpretation

Gemini AI interprets the learner’s goal and generates an initial roadmap blueprint.

Example goals:

Become a Full-Stack Engineer
Transition into Product Management
Improve Emergency Triage skills

The model extracts key skill concepts and generates the initial roadmap structure.

3. Skill Graph Construction

Goals are mapped to a skill dependency graph.

Each node represents a competency and includes:

prerequisite relationships
target mastery levels
learning concepts
checkpoint templates

This allows the system to understand how skills build on one another, rather than treating learning as a flat list.

4. Skill Gap Analysis

The learner’s baseline skills are compared against the target graph.

Gap severity is calculated as:

$$ gap = max(0, target_skill - learner_skill) $$

Additional signals include:

prerequisite urgency
bottleneck risk
downstream dependency impact

5. ML Recommendation Engine

Candidate roadmap nodes are ranked using a logistic regression model trained on synthetic learning data.

The ranking function:

$$ P(node) = \sigma(w_0 + w_1x_1 + w_2x_2 + ... + w_nx_n) $$

Where features include:

gap severity
pace deviation
timeline pressure
success likelihood
bottleneck relief potential

Example simplified scoring logic:

function scoreNode(features, weights) {
  let z = weights.intercept;

  for (let i = 0; i < features.length; i++) {
    z += features[i] * weights[i];
  }

  return 1 / (1 + Math.exp(-z));
}

The highest scoring node becomes the next recommended learning step, along with an explanation of why it was selected.

6. Practice Checkpoints

Each roadmap node includes a practical checkpoint designed to simulate real work tasks.

Examples include:

coding challenges
project exercises
written analysis
applied real-world scenarios

These checkpoints ensure learners demonstrate true skill development rather than passive consumption.

7. Bayesian Mastery Tracking

Skill mastery is tracked using a Beta-Binomial Bayesian model.

Skill probability:

$$ mastery = \frac{\alpha}{\alpha + \beta} $$

Checkpoint evidence updates the distribution:

alpha += performance * evidence
beta  += (1 - performance) * evidence

Signals used in updates:

checkpoint score
learner confidence
retries
time spent

This provides both mastery estimation and confidence intervals.

8. Dynamic Roadmap Adaptation

After every checkpoint:

skill mastery is updated
pacing signals are recalculated
bottlenecks are detected

The recommendation model then re-ranks roadmap nodes to determine the next optimal step.

This creates a continuous learning feedback loop.

9. Learning Analytics Dashboard

A React + Tailwind + Recharts frontend visualizes learner progress through:

skill mastery maps
roadmap progress
pacing trends
bottleneck detection
readiness scores

This allows learners to clearly see where they stand and what to focus on next.

10. Organization & Cohort System

SkillPath also includes multi-user organization workspaces.

Organizations can:

create a team workspace
invite learners via email
assign shared learning goals
monitor individual and cohort progress

The backend includes models for:

Organization
Membership
LearnerProfile
GoalPlan

Each learner still receives a personalized roadmap, while the organization dashboard tracks roadmap completion progress across members.

This allows instructors, bootcamps, and companies to monitor how learners are progressing toward shared goals.

Challenges we ran into

Designing meaningful learning signals- simple completion metrics were not enough. We modeled retries, confidence, time spent, and checkpoint outcomes to estimate mastery.
Adaptive roadmap generation- traditional courses are linear. Our system required dynamic dependency graphs where nodes unlock based on learner performance.
Cold start problem- new learners have little data. Bayesian priors allow reasonable skill estimates early on.

Accomplishments that we're proud of

Built a full learning intelligence pipeline that continuously analyzes learner data.
Generated dynamic personalized roadmaps where two learners with the same goal receive different paths.
Built a real-time analytics dashboard visualizing mastery, pacing, and bottlenecks.
Implemented an end-to-end feedback loop where every checkpoint reshapes the roadmap.
Built an organization and cohort tracking system that allows teams to invite learners and monitor skill development across the group.

What we learned

Feedback loops are essential - recommendations must evolve as learners progress.
Feature engineering matters more than the model - designing the right learning signals had the biggest impact.