Description: An enterprise extension of the LINX research: a pre-enrollment identity layer for LinkedIn Learning that gives L&D managers a diagnosis, not just data. Built to demonstrate full product analyst skills: stakeholder personas, MoSCoW prioritisation, Agile sprint planning, UML diagramming, and an interactive Figma prototype.

The Problem: Companies spend $1,254 per employee on training. Only 8% of CEOs see the business impact. The gap isn't budget - it's identity. L&D managers have completion data but no explanation for why employees abandon.

What I Built:

• PRD with personas (Monica, Ramanik, Vanya), 10 MoSCoW features, user stories, and 30/60/90 success metrics
• Product lifecycle diagram: Discovery → Define → Design → Build → Launch → Measure
• UML use case and activity diagrams for three actors: Line Manager, Employee, L&D Manager
• Agile Sprint 1 plan with epics, story points, task breakdown, and definition of done
• Interactive Figma prototype: 3 flows, 18 screens, CloudZZA fictional enterprise client

Key Design Decision: A Mandatory / Good to Have toggle on course assignment determines whether LINX activates — one UI decision that encodes the entire product philosophy.

Read the full article on Medium.

Explore the Figma Prototype · PRD

Description: A redesign of LinkedIn Learning's onboarding flow, built directly from the findings of Lost in the Learning Loop. Introduces LINX - LinkedIn's Learning and Identity Experience: a feature that asks who you're becoming before recommending what to learn.

Key Findings: The learning decision is an identity decision in disguise. The variable separating commitment from drift was not discipline or time — it was clarity about who the learner was becoming.

• LinkedIn Learning's onboarding assumes the identity decision is already made; for most learners it isn't
• Three identity paths replace skill selection, each mapped to a research archetype
• LINX uses LinkedIn's existing connection graph to personalise recommendations without new data
• Commitment Echo re-anchors identity commitment at Module 2, the exact drift point identified in research

Grounded in:20-person survey, 4 moderate interviews, journey map systhesis, four learner archetypes.

Tools: UX Research, Figma Prototype, Research Grounded

Read the full case study on Medium.

Explore the Figma Prototype

Description: How do career-driven learners decide what to study next — and why do 75% abandon what they start? An independent mixed-methods study combining a quantitative survey with four moderated interviews, conducted January–March 2026.

Key Findings: The learning decision is an identity decision in disguise. The variable separating commitment from drift was not discipline or time — it was clarity about who the learner was becoming.

• 75% of respondents abandoned at least one course — the top reason wasn't losing motivation, it was finding something more interesting
• Excitement, pressure, and overwhelm were felt simultaneously — not separately
• Four learner archetypes emerged: Decided, Searching, Drifting, and Moving
• The variable separating commitment from drift was not discipline or time — it was identity clarity
• Core finding: the learning decision is an identity decision in disguise

Design Recommendations:

1. Redesign the onboarding question from "what do you want to learn?" to "who are you becoming by choosing this?"
2. Treat the entry decision as a design surface, not a friction point to eliminate
3. Distinguish between stuck and blocked learners — they need different interventions

Tools: Fillout (survey), Otter.ai (transcripts), FigJam (affinity mapping), mixed-methods thematic analysis

Read the full case study on Medium.

View the affinity maps on FigJam

Description: Analysed 32,593 Open University learners to uncover what drives student withdrawal — and what to do about it.

Key Insights:

• 10,156 learners disengaged between enrolment and active participation
• No formal qualifications = 42.9% withdrawal rate (highest of any group)
• Youngest learners (0-35) churn at 32.2% vs 25% for 55+
• Course CCC anomaly: average scores but highest dropout; course design drives disengagement more than ability
• Early assessment score alone does not predict withdrawal (FFF outlier)

Three Priorities for Retention:

1. Trigger engagement check-in at week 2 of enrolment
2. Audit course design for CCC, not student ability
3. Targeted support for 0-35 age group

Tools: Python, Tableau, OULAD Open Dataset (Open University)

Checkout the work at Tableau.

Description: A comprehensive SQL analysis project exploring library operations through data-driven insights.

Built a relational database with 3 tables (Books, Members, Checkouts) and performed 13 analytical queries to answer real-world business questions about circulation patterns, member engagement, and revenue tracking.

Key Insights:

• Analyzed 25 books and 30+ checkout transactions to identify high-demand titles and popular genres
• Calculated $45.79 in late fee revenue using CASE statements and advanced date functions
• Discovered Fantasy and Dystopian genres drive 40% of total circulation
• Identified 96% member engagement rate with only 1 active member never borrowing
• Used window functions to rank members by activity and track cumulative checkout trends

Technical Highlights:

• Complex JOINs (INNER, LEFT, OUTER) to combine data across multiple tables
• Window functions (RANK, DENSE_RANK, SUM OVER) for rankings and running totals
• Subqueries and aggregate functions (COUNT, AVG, SUM, ROUND) for calculations
• CASE statements for conditional logic in late fee calculations
• Date manipulation using julianday() to solve TEXT-to-DATE conversion challenges

Technologies: SQLite, SQL (Advanced queries), dbdiagram.io

For the complete code, visit the Git repo.

Analyzed 17 years of economic data to test whether Germany's renewable energy policy caused industrial decine.

Combined Eurostat price data with production indices across 4 energy-intensive sectors (chemicals, metals, glass, paper) to quantify the relationship between electricity costs and manufacturing output.

Key Findings:

• Industrial electricity prices doubled from €0.115/kWh to €0.231/kWh (2007-2024)
• Energy-intensive manufacturing declined by average of 22.4% across all sectors
• Regression analysis revealed electricity costs explain only 13-19% of production decline
• 2022 energy crisis (market shock) caused more damage than 15 years of policy surcharges
• Metals sector showed strongest price-production relationship (r = -0.44, R² = 0.19)
• Statistical analysis contradicts simple causation: multiple factors drive industrial decline

Technical Highlights:

• Time-series analysis with 17-year dataset (37 semi-annual periods)
• Statistical modeling: correlation analysis, linear regression, R² interpretation
• Python libraries: pandas, matplotlib, scipy, statsmodels
• Data integration from multiple sources (Eurostat, Destatis)
• Hypothesis testing with nuanced interpretation of results

Policy Implications:

Germany's experience demonstrates that energy security and price stability matter more than gradual policy-driven cost increases.

2022 crisis occurring after the EEG renewable surcharge was abolished—proves geopolitical risk management is as critical as climate policy design.

Technologies: Python, Pandas, Matplotlib, SciPy, Statsmodels, Jupyter Notebook

For the complete code, visit the Git repo.

A comprehensive machine learning project analyzing thyroid cancer risk factors using logistic regression and Random Forest models.

Built predictive models to identify key risk indicators and achieved 71.74% accuracy on balanced data.

Project Highlights:

• Analyzed 3,700+ patient records with 13 clinical and demographic variables
• Identified 3 critical risk factors: Radiation Exposure, Iodine Deficiency, and Family History (p < 2.2e-16)
• Achieved 71.74% accuracy using Random Forest with balanced dataset
• Addressed severe class imbalance using random under-sampling techniques
• Developed both binary (Benign/Malignant) and multi-class (Low/Medium/High risk) classification models
• Conducted comprehensive EDA with Chi-Square tests and visualization analysis

Technical Stack:

• R Programming (ggplot2, caret, randomForest packages)
• Statistical Analysis (Chi-Square tests, logistic regression, multinomial regression)
• Machine Learning (Random Forest, model evaluation, confusion matrices)

Critical Insights:

• Radiation exposure, iodine deficiency, and family history are strongest predictors of malignancy (p < 2.2e-16)
• Age and gender showed no significant association with cancer risk
• Class imbalance (77% benign) initially biased models toward majority class
• After balancing: Logistic Regression achieved 89.44% sensitivity with 100% specificity
• Random Forest showed 89.44% sensitivity but lower specificity (65.04%)
• Multi-class model struggled with Medium-risk category (0% sensitivity)

Built with: Tableau Public | Statistical Modeling | Strategic Analysis

For the complete code, visit the Git repo.

The $50M Question: Where Can You Actually Hire?

When a Fortune 500 CSO needed to choose between Healthcare and Construction for a major expansion, the question wasn't the market size, it was execution feasibility.

This analysis of 126 months of Bureau of Labor Statistics data revealed:

• September 2022 marked a fundamental labor market break
• Healthcare shows 52% job posting failure vs. Construction's 12%
• Three industries remain in structural shortage 35+ months later
• Hiring in Healthcare is 2.3x harder than Construction

Custom metrics developed:

• Talent Acquisition Pressure (months per hire)
• Investment Risk Score (0-100 difficulty index)
• Shortage Classification (structural vs. cyclical)

Built with: Tableau Public | Statistical Modeling | Strategic Analysis

Check out the Simulator yourself at Tableau.