Business Analyst with 3+ years in energy & digital transformation at Schneider Electric. Now combining that commercial instinct with an MSc in Business Analytics at Trinity College Dublin.
I started my career not in a data lab, but in the field — managing government accounts worth hundreds of crores, navigating complex stakeholder ecosystems at BHEL and NTPC, and making the case for digital energy solutions to engineers and executives who needed more than a sales pitch. They needed evidence.
That pressure taught me something no classroom could: data without business context is just noise. Every dashboard I built, every market analysis I ran, every KPI I tracked had to justify itself in hard commercial terms — deal velocity, retention rates, margin impact.
The turning point came during the evaluation of India's Green Hydrogen opportunity — a ₹500Cr+ market I had to make legible for five internal teams simultaneously. I realised that the gap between good strategy and great strategy is almost always a data infrastructure problem. That's when I decided to formalise my analytical skills.
Now at Trinity College Dublin, I'm learning to build the models and systems that I always needed but had to work around. My edge is rare: I've sat on both sides of the insight. I know what a VP actually needs from a dashboard, and I know how to build it.
How do you unlock $28M in value from 40M patient records when 78% of the data is trapped in unstructured text across three incompatible post-acquisition architectures?
Designed a 6-layer Medallion Lakehouse (Bronze-Silver-Gold) on Azure Databricks to consolidate three incompatible post-acquisition architectures (Paige AI, Ambry Genetics, Deep6 AI). Built an Agentic NLP pipeline using ClinicalBERT and Bio-BERT to extract structured clinical entities at scale, with MLflow governance and RAG infrastructure.
Only 22% of 40M patient records were AI-ready. The bottleneck was not data volume but schema incompatibility and unstructured text. A confidence-gated multi-agent NLP pipeline (≥0.90 auto-accept; 0.70–0.89 HITL review) enabled scalable extraction while preserving clinical accuracy and audit provenance.
Roadmap to increase structured data coverage from 22% to 50% within 18 months. Projected 3-year NPV of $28M, reducing manual abstraction cost from $38 to under $12 per chart. Enabled pharmaceutical-grade audit trails for Pfizer, AstraZeneca, and GSK partners.
With 519,409 reviews and severe class imbalance, can we identify exactly which customer segments drive negative sentiment — and what Sephora should do about it?
Compared three NLP models — Naïve Bayes baseline, Logistic Regression + TF-IDF, and fine-tuned BERT (bert-base-uncased, 110M params, 3 epochs). Applied LDA topic modelling to surface 7 complaint themes, and segmented results by skin type, skin tone, and price tier.
BERT outperformed LR by +26pp on negative F1 (0.76 vs 0.51). Darker skin tones showed a 67% higher negative review rate vs fair/porcelain. Dry skin + budget products had the worst dissatisfaction (12.3%). La Mer hit 19.8% negative despite premium pricing — a value perception problem, not a quality problem.
4 actionable recommendations: reformulate for dry/sensitive skin, expand shade range for deeper tones, address value-for-money perception with sample sizes and outcome messaging, and deploy BERT for real-time review monitoring at scale.
How can a non-technical business user get trend analysis, risk flags, KPIs, and strategic recommendations from their own data in seconds — without writing a single line of code?
Built a full-stack AI analytics application in Python + Streamlit. Users drag-and-drop a CSV, XLSX, or XLS file (up to 200MB), select their industry and report type (Executive Summary, Risk Analysis, Trend Report), and InsightIQ uses a Groq-powered LLM (Llama 3.2 11B) to generate structured analysis and interactive visualisations. Supports Markdown, Word, and PDF export.
Designed a 4-step UX: Upload → Analyse → Ask → Decide. The "Ask AI about your data" chat layer allows free-text business questions against the uploaded dataset — functioning like a junior analyst on demand. Industry-aware prompting adapts recommendations to context (retail, finance, healthcare, etc.).
Reduces time-to-insight from hours to under 60 seconds. Designed for non-technical decision-makers — no SQL, no Python required. Demonstrates end-to-end product thinking: UX design, LLM integration, export functionality, and real business framing — not just a notebook.
Why are 20,000+ transactions failing — and which failures actually cost the business?
End-to-end process mapping on 20,000+ electronics transactions. Chi-square testing validated statistical significance of failure patterns; Pareto modelling prioritised fixes by ROI impact. Customer risk segmentation identified high-value loss cohorts.
Top 3 cancellation drivers responsible for 80% of revenue leakage. Failure points mapped to specific process stages, enabling surgical corrective action rather than blanket fixes.
Delivered BA-style recommendations with Pareto-driven ROI modelling — directly replicating root cause and process improvement methodology used in enterprise analytics environments.
Which customers are about to leave — and what would actually change their mind?
Multi-model churn prediction pipeline using XGBoost and Random Forest with SHAP explainability to surface the 'why' behind predictions, not just the score. Mirrors customer-facing analytics used in energy retail contexts like Electric Ireland.
SHAP revealed which customer behaviours were the most powerful churn predictors — translating black-box outputs into actionable, prioritised retention signals for non-technical stakeholders.
Delivered business-ready retention recommendations structured for executive consumption, with intervention priority ranked by predicted revenue risk.
Can we identify at-risk students early enough to intervene — before it's too late?
Logistic Regression and Decision Tree models applied to 6,378+ student records at Trinity College Dublin. Built in R with full model validation pipeline to predict academic risk before formal assessment checkpoints.
Model achieved 90.49% accuracy. Decision tree visualisation allowed non-technical academic staff to understand risk logic directly — making findings actionable beyond the data team.
Enabled an early-identification framework for targeted academic support — a meaningful institutional shift from reactive intervention to proactive student success management.
With 278,000+ IMDb records — what actually makes a film commercially viable?
Data-backed creative strategy analysis using 278,000+ IMDb records (2005–2025). Python for data wrangling and EDA; Tableau for an interactive dashboard targeted at a new film production venture's leadership team.
Identified optimal genre mix, ideal runtime window of 95–110 minutes, and talent positioning patterns correlated with maximum commercial ROI — patterns studios overlook when relying on gut instinct alone.
Delivered an executive-ready strategy deck moving creative decision-making from instinct to data-validated recommendations a new production company could act on immediately.
How do you give a national sales team real-time visibility — without drowning them in data?
Comprehensive KPI dashboard for Schneider Electric India's P&G segment using Excel and Tableau. Built backwards from what sales managers actually need to make weekly decisions — not exhaustive reporting for its own sake.
Pipeline velocity and account engagement rate identified as leading KPIs predicting quarterly performance up to 6 weeks ahead — shifting the team from reactive to anticipatory management.
Drove a 20% improvement in target achievement efficiency. Dashboard became the standard reporting tool for the segment, replacing fragmented spreadsheets with a single source of truth.
Every analysis I run follows the same discipline — rooted in commercial reality, not just technical execution.
Business problems are rarely what they first appear. I spend time with stakeholders to separate symptoms from root causes before touching any data.
What data exists? What's missing? What are the quality risks? I document assumptions and constraints upfront — it saves hours downstream.
Every model, dashboard, or analysis is built backwards from the decision it needs to support. Insight without action is just a report.
I translate technical outputs into clear, quantified recommendations that non-technical stakeholders can act on — immediately.
I'm actively seeking Business Analyst, Data Analyst, and Analytics Consultant roles in Dublin and across Ireland. Open to conversations about where analytical thinking meets real business impact.
Send a message →