Understanding the problem

Across global job platforms, one pattern is consistent:

People do not lack ambition — they lack momentum.

We uncovered three critical gaps:

How AI + Behavioural Science Solves It

CareerPro combines machine intelligence with human behavioural insight to create a companion that moves job seekers forward, step by step.

1. AI Understands the Job Seeker’s Context: Our models analyze behavioural patterns, profile data, and job market trends to build a dynamic understanding of each user’s journey.

2. Behavioural Science Predicts What They Need Next: We use proven behavioural principles to detect when a person is likely to stall—then deliver the precise nudge or micro-task that re-engages them.

3. Real-Time Micro-Guidance Reduces Friction: Instead of overwhelming checklists, CareerPro offers bite-sized tasks, timely nudges, personalized recommendations, and structured steps toward a goal.

4. Continuous Learning Makes the Experience Smarter: The more CareerPro interacts, the more it adapts ensuring every suggestion is contextual, relevant, and actionable.

We are passionate professionals committed to delivering exceptional results that exceed expectations and drive meaningful transformation.

Step-by-Step Flow

1. Create Your Profile: CareerPro walks you through a clean, step-by-step setup that captures your experience, skills, education, and goals without the overwhelm.

Each session is supported by smart prompts that help you optimise your actions for success.

2. Set Your Goal: Apply to desired Roles, Skills & Industries. Define the direction you want to grow. CareerPro aligns your profile with specific job titles, required skills, and relevant industries to create a personalized nudges. Your journey is more than data it’s personal. CareerPro guides how you interact with the job seeking platform based on what actions will optimize your success, including:

This allows the system to offer guidance tailored to you.

3. AI-Powered Action Guidance
With Micro-Tasks instead of giant to-do lists, CareerPro gives you small, achievable steps write a line, update a skill, apply to one job because it is designed to build momentum and reduce overwhelm. With optimised nudges – if the system sees you slowing down or getting stuck, it gently nudges you at the right moment. These prompts are grounded in behavioural science to help you progress without pressure.

Support keeps you moving forward, one step at a time.

4. Track Your Progress
Using dashboard analytics see your job search come to life with a real-time dashboard showing your activity, readiness, and progress toward your goals. By having employer engagement insights, learn when and how employers engage with your profile.

CareerPro helps you refine your job seeking strategy, improve response rates, and stay proactive.

System Design and Architecture

• Frontend: We used Streamlit as our frontend framework. It is python based, and easy to integrate with FastAPI and ML endpoints.
• Deployment: We deployed our app to google cloud for public access.
• Backend: We had an MCP backend server with FastAPI endpoints. For some requests it will go through our SQList database for data storage and retrieval. For requests that require an LLM agent, the MCP server will determine on its own which APIs to call, and then it will integrate all the information to send to the Claude LLM agent for advanced reasoning and content generation.
• Data pipeline & Modeling: We majorly used SageMaker Studio for our notebooks, including EDA, feature engineering, model training and deployment. We also store our raw data as well as all the checkpoints between different notebooks in S3. After the model is deployed, it will serve as a SageMaker real-time endpoint, and the frontend will call the endpoint to make inferences.

Data Pipeline

1. Data Ingestion Layer Sources: CVs, user profiles, job descriptions, behavioural logs, employer feedback Cleaning & normalization PII stripping and anonymization

2. Feature Engineering Skills extraction (NER) Experience relevance scoring Behavioral markers (drop-offs, hesitations, inactivity windows) Context cues (application stage, job domain, time-of-day engagement)

3. ML Models Matching Model: Rank job–profile fit using embeddings Behavioural Prediction Model: Predict drop-off risk, nudge type Skill Gap Model: Suggest learning interventions Recommendation Engine: Micro-tasks Job insights Employer-fit score Optional: LLM-driven guidance text generation

4. Real-Time Decision Layer Determines when to send nudges Generates personalized messages

5. Evaluation Metrics MSE, RMSE of the model accuracy against user actions behavioural uplift metrics

Model Selection

Our modeling goal is to predict which user behaviors will increase engagement outcomes. Specifically, we’re looking at outcomes like Candidate Profile views and Employer reach. We evaluated the following models:

1. Linear Regression: A linear regression model was used to establish a clear baseline as it is fast to train and helps measure whether a more complex model adds value. We used it to set the minimum performance bar (R² value). 

2. Neural Networks: We used neural networks as they are capable of learning complex, non-linear relationships that simple models can’t capture and because they are good for large datasets such as the one we are using. Neural networks provided substantial improvement over the baseline, capturing non-linear relationships but required heavy tuning and were less interpretable. 

3. Random Forest: Random forest handles nonlinear patterns well and is a good model to identify strong signals in noisy, behavioural data. It performed consistently across targets and gave us a reliable, benchmark performance for nonlinear models. 

4. XGBoost: XGBoost excels at modeling non-linear relationships and feature interactions. It is efficient, scalable, highly tunable and handles sparse features well. Since our dataset consisted of skewed and high-variance behavioral signals with sparse feature distributions, we used XGBoost for its accuracy, generalization and feature interpretability (via Shap). We found that XGBoost consistently outperformed other models and was the best choice for production as it balanced speed, performance and transparency. Feature importance and SHAP explanations aligned well with real behavioural assumptions. 

Model Evaluation

After comprehensive testing, we selected XGBoost as our final model as it consistently outperformed the other models for both target outcomes.
It captured non-linear patterns across multiple time scales without overfitting. And it provided the interpretability we needed through SHAP analysis.

SHAP Analysis

The critical piece is the SHAP analysis at the individual user level as it
reveals which features actually drive outcomes. It transforms model predictions into personalized nudges like ‘login more frequently’ or ‘apply to more jobs’ which we use to tell the user exactly what to do next to increase engagement outcomes. 

Today’s job seekers face overwhelming friction

• 1 in 3 abandon applications
• 6 in 10 feel unsupported
• Majority ignore email nudges

CareerPro delivers guidance exactly when you need it under 3 pillars:

• Contextual AI
• Behavioural Insights
• Real-time Nudges

We help you increase engagement, improve application success rates, and land your dream job faster.

Testimonials

“The AI’s utility in suggesting profile improvements and guiding the skill-adding process is wonderful.“

” The AI successfully identified beneficial skills to add to a user’s job search.“

We deliver exceptional results through proven expertise, cutting-edge innovation, and unwavering commitment to your success. Our comprehensive approach ensures sustainable growth and competitive advantage.

Our Vision & Mission

Our Vision is a world where every job seeker gets personalized, real-time support.

Our Mission is to make job seeking a confident, guided, and rewarding experience.

Our Story

Finding a job should feel hopeful. But for millions of people, it feels overwhelming, lonely, and confusing. CareerPro was born from a simple observation: people are not failing at job seeking, the system is failing them.

We spoke to job seekers who were motivated, skilled, and eager to grow. Yet most of them were stuck—not because they lacked talent, but because they lacked the right guidance at the right moment. Their journey wasn’t broken. It just wasn’t supported.

CareerPro grew out of that insight: to turn job seeking into a guided, encouraging, and ultimately empowering experience.

At the same time, hiring teams told us they struggled to find prepared, motivated candidates because the job search funnel was full of drop-offs and incomplete profiles. We realized the friction wasn’t individual it was systemic. That’s when the idea of CareerPro emerged: an assistant that stays with you through the entire journey, not just at the start or the end.

Our Values

Accessibility

We believe opportunity should never be limited by background, geography, or circumstance. CareerPro is built to make high-quality career guidance available to every job seeker, not just the privileged few. From intuitive design to inclusive language and low-bandwidth access, we ensure that everyone can take their next step with confidence.

Empowerment

Job seekers deserve more than information they deserve the tools, support, and momentum to move forward. CareerPro empowers individuals with actionable guidance, clear pathways, and personalized insights that help them own their journey, build their confidence, and unlock new possibilities.

Transparency

Trust is the foundation of meaningful career support. We are committed to being open about how recommendations are made, how data is used, and how decisions are shaped. CareerPro provides clear explanations and honest insights, ensuring users always understand the “why” behind every suggestion.

Human-Centered AI

Technology should amplify human potential not replace it. Our AI is designed around real needs, real behaviours, and real aspirations. We combine ethical AI practices with behavioural science to ensure every interaction feels relevant, respectful, and deeply human. CareerPro adapts to users, not the other way around.

---

The team