I work across the full analytics stack — translating raw, messy data into decisions that drive business outcomes.
| Area | What I Deliver |
|---|---|
| 📊 Product & Business Analytics | Funnel analysis, A/B experimentation, user segmentation, churn modeling |
| 📣 Marketing & Campaign Analytics | Incrementality testing, ROI measurement, A/B significance pipelines, influencer ROI |
| 🗄 SQL & Data Engineering | Complex queries (CTEs, window functions, JOINs), ETL pipelines, PySpark, Hive |
| 📈 BI & Dashboards | Self-service Tableau / Power BI dashboards built for business stakeholders |
| 🤖 AI & LLM Systems | LLM-powered reporting automation, agentic workflows, n8n pipelines |
Currently open to
Product Analytics/Business Analytics/Marketing Analyticsroles
| 📊 15,000+ | 💰 18.2% | 🛒 40–63% | 🤖 6+ |
|---|---|---|---|
| UPI transactions analyzed & fraud-patterned | Gender pay gap quantified across 8K Indian tech records | Delivery fee surge exposed on rain days | LLM agents built end-to-end |
| ⚡ 13M+ | 📉 51pp | 🎯 137.8% | 🕐 30s |
|---|---|---|---|
| Incrementality test records processed in PySpark | Retention gap exposed after a single failed fintech transaction | Portfolio ROI on influencer campaign tracker | Executive reporting cycle cut from 2–4 hrs to 30 sec with LLM |
Languages & Query
Analytics & BI
Data Engineering
ML & AI
AI Dev Tools
- Marketing & Campaign Analytics
- Product Analytics
- Data Engineering
- SQL
- Python & EDA
- Machine Learning
- AI & LLM Agents
- Tableau & Dashboards
Business Problem: How do you measure what a campaign actually caused — not just what happened during it?
End-to-end marketing analytics pipelines: incrementality testing, A/B significance at scale, influencer ROI, and LLM-powered reporting — built with PySpark, Hive, pandas, and Groq.
| Project | Problem Solved | Tools | Key Result |
|---|---|---|---|
| 📈 Acquisition Campaign Incrementality & ROI | Separate true campaign lift from baseline conversion noise across 13M+ test records | PySpark, Hive, chi-square, t-test | Ranked targeting segments by verified incremental ROI; isolated statistically significant lift vs organic baseline |
| 🧪 A/B Test Statistical Significance Pipeline | Run concurrent marketing experiments and auto-rank results: scale, stop, or inconclusive | PySpark, chi-square, t-test, logistic regression | 9/10 experiments reached significance; EXP_005 showed a negative effect (−0.4pp) flagged for immediate stop |
| 🤖 AI Business Reporting & Sentiment Analysis | Cut 2–4 hour manual reporting cycles down to 30 seconds using LLM automation | Python, pandas, Groq (Llama 3.3 70B), n8n | Executive summary + VoC sentiment report auto-generated from raw sales CSV; 40% positive, 30% negative sentiment surfaced |
| 📸 Influencer Campaign Performance Tracker | Measure reach, engagement, and ROI across 40 influencer partnerships and flag who to scale or drop | Python, pandas, Groq LLM, matplotlib | 137.8% portfolio ROI, 1.65x ROMI; Groq LLM auto-flagged 11 SCALE and 12 DROP partnerships |
Business Problem: Where in the user journey are you losing people, and what does the data say you should fix?
Quantitative funnel analysis combined with structured qualitative synthesis — demonstrating how product analysts translate user pain points into prioritised, data-backed recommendations.
| Project | Problem Solved | Tools | Key Finding |
|---|---|---|---|
| 💳 Fintech User Funnel Analytics | Identify where a youth payments app (ages 11–19) loses users and which transactions fail most | Python, pandas, matplotlib, seaborn | 51pp retention gap after a failed first transaction (92% vs 41%); first-time UPI failure rate 30.7% vs 9.1% repeat; 3 prioritised product recommendations with success metrics |
Business Problem: How do you reliably move, transform, and serve data at scale without manual intervention?
Automated pipelines built with Airflow, PostgreSQL, Docker, and FastAPI — containerised and production-ready.
| Project | Problem Solved | Tools | Key Result |
|---|---|---|---|
| 🌦 Weather Data ETL Pipeline | Manual weather data collection was slow and error-prone | Python, Airflow, PostgreSQL, Docker, Open-Meteo API | Fully automated daily ingestion pipeline — zero manual effort, containerised for one-command deployment |
| 🧥 Apparel Analytics Hub | Apparel retail data sat siloed with no aggregated reporting | Python, Airflow, Flask | Modular DAG-driven ETL serving aggregated retail metrics through a Flask web app |
| 🧠 AI Context Engine | AI conversation context was lost between browser sessions | FastAPI, Chrome Extension, WebSockets | Full-stack extension capturing and restoring full AI session context across sessions |
Business Problem: How do you extract decision-relevant signals from complex, multi-table datasets?
Advanced SQL across fraud detection, pay equity, consumer pricing, and job market analysis — using CTEs, window functions, and complex JOINs.
| Project | Problem Solved | Approach | Key Finding |
|---|---|---|---|
| 💡 8-Week SQL Challenge | Demonstrate SQL depth across 8 real-world business scenarios | CTEs, window functions, cohort analysis, KPI reporting | Solved all 8 case studies from #8WeekSQLChallenge |
| 💼 Job Market Analytics | Which skills, geographies, and salary bands have the highest demand? | YoY/MoM trend comparisons, window functions | Tracked skill demand shifts with automated Tableau refresh |
| 💳 UPI Fraud Analysis | What transaction patterns predict fraud in UPI payments? | 5 anomaly detection rules across 15,000 transactions | Late-night transactions: 19.2% fraud rate vs 0.8% mornings; amounts near ₹4,999/₹9,999 show 28–31% fraud rates |
| 💰 Indian Salary Gap Analysis | How large is the gender pay gap in Indian tech, and what drives it? | Segmentation by role, city, company type across 8,000 records | 18.2% gender pay gap — Bangalore commands 32–35% city premium; MNCs pay 40.3% more than Indian corporates |
| 🍔 Swiggy Dark Patterns | Are food delivery platforms using pricing dark patterns against consumers? | Controlled comparisons isolating weather, time, and demand as pricing levers | 40–63% delivery fee surge during rain; weekend fees 24.9% higher — costing users ₹864/year |
Business Problem: What patterns, anomalies, and segments hide in raw data that descriptive stats alone won't surface?
Exploratory analysis tackling user behavior, competitive pricing, workforce analytics, and environmental prediction.
| Project | Problem Solved | Approach | Key Insight |
|---|---|---|---|
| 💳 Fintech User Funnel Analytics | Where does a youth fintech app lose users, and which segments churn hardest? | Funnel drop-off analysis, cohort split (11–14 vs 15–19), failure-rate segmentation | 51pp retention cliff after failed first transaction; age 15–19 transacts at 2.3× the value of 11–14 cohort |
| 📸 Influencer Campaign Tracker | Which of 40 influencer partnerships are worth scaling, and which should be cut? | ROI, ROMI, CPA, engagement rate computed per influencer; LLM-generated recommendations | Identified 11 SCALE and 12 DROP creators; TikTok macro tier at 8.3x ROMI, Instagram macro at sub-1x |
| 🌱 HR Analytics | What drives employee attrition, and where are the pay inequities? | Segmented 8,950 employees by education, department, and performance | Education strongly predicts earnings: high school ₹62K avg → PhD ₹92K; attrition clustered in specific department-seniority bands |
| 🛒 Zepto vs Blinkit | Which quick-commerce platform offers better value, and where does each dominate? | Compared pricing, discounts, delivery times across 5,000+ products | Zepto 11.8 min vs Blinkit 16.5 min delivery; ₹34.94 vs ₹17.52 avg discount — picked up 6,000+ LinkedIn views |
| 💳 UPI Fraud Analysis | Which user behavioural patterns signal fraudulent UPI activity? | Built 5 rule-based fraud signals from 15K transactions | Time-of-day, recipient novelty, retry behaviour, and amount clustering are strongest fraud predictors |
| 💰 Indian Salary Gap Analysis | How do role, seniority, city, and company type compound the pay gap? | Statistical breakdown across 8,000 Indian tech compensation records | MNCs pay 40.3% more; seniority and city amplify the gender gap significantly |
| 🍔 Swiggy Dark Patterns | How much extra are consumers paying due to platform-driven pricing manipulation? | Isolated weather, time-of-day, and demand as independent pricing levers | r = −0.71 correlation between delivery time and satisfaction; users overcharged ₹864/year via opaque fees |
| 🌫 AQI Prediction | Can air quality be predicted spatially from vegetation and geography data? | Spatial interpolation across geographic regions | Interactive HTML map visualising AQI vs vegetation — surfaced high-risk zones |
Business Problem: Can we build predictive systems that act on patterns too complex for manual rules?
End-to-end ML projects covering classification, regression, NLP, computer vision, and deployed systems.
| Project | Problem Solved | Model / Approach | Result |
|---|---|---|---|
| 🤖 ML Portfolio | Demonstrate end-to-end ML across 10 diverse business domains | Logistic regression, CNN, YOLO, TF-IDF, collaborative filtering | 10 complete projects: churn, vision, NLP, forecasting, recommendation |
| 🔐 User Authentication ML System | How can we detect anomalous login behaviour in real time? | Decision tree + JWT auth, deployed on Render + Netlify | Production-deployed full-stack app with real-time ML inference |
| 🏨 Hotel Booking Prediction | Which bookings are likely to cancel, and when? | Logistic regression + tree-based models on booking records | Enabled proactive demand management by predicting cancellation probability |
| 🎬 Movie Recommendation System | How do you personalise content recommendations at scale? | TF-IDF vectorisation + cosine similarity on plot/genre metadata | Content-based recommendations without user history dependency |
| 📰 Fake News Predictor | Can NLP distinguish misinformation from legitimate news? | TF-IDF + ML classifier on labelled article datasets | Classifier trained to detect misinformation across news topics |
| 📺 YouTube Ads View Prediction | Which engagement signals best predict ad view volume? | Regression on likes, comments, shares | Quantified the relationship between engagement metrics and ad performance |
| 📷 Real-Time Object Detection | Can we detect multiple object classes in a live video stream? | YOLOv8 with custom YAML-defined classes | Real-time detection pipeline with configurable object categories |
Business Problem: Which repetitive analytical and reasoning workflows can be handed off to autonomous AI agents?
Agentic systems built with CrewAI, OpenAI, Groq, and FastAPI — developed using Cursor and Claude Code.
| Project | Problem Solved | Tools | Key Outcome |
|---|---|---|---|
| 📊 AI Business Reporting & Sentiment | Auto-generate executive summaries and VoC sentiment reports from raw sales + review CSVs | Python, pandas, Groq Llama 3.3 70B, n8n | 2–4 hour manual reporting cycle → under 30 seconds; real Groq output committed as sample evidence |
| 📸 Influencer Campaign Tracker | Auto-flag 40 influencer partnerships as SCALE / MONITOR / OPTIMIZE / DROP using live campaign data | Python, pandas, Groq LLM, matplotlib | Portfolio ROI 137.8%, 1.65x ROMI; LLM-generated strategic recommendations with data grounding |
| 🧠 LLM Portfolio | Build a diverse suite of production-grade agentic AI systems | CrewAI, OpenAI, Gemini, HuggingFace, FastAPI, Gradio | 6 agents: medical vision+voice, finance analyst, image recognition, symptom analysis |
| 🩺 Blood Test Analyser | Automate interpretation of blood test PDFs into plain-English summaries | CrewAI, OpenAI, Python | Multi-agent system reading PDF biomarkers and generating patient-friendly medical summaries |
| 📰 Financial News Dashboard | Surface the most relevant financial news from a high-volume feed | Python, Streamlit, FAISS, Embeddings | Semantic similarity search over embedded news articles |
| 📝 Text Summarizer | Automate extractive and abstractive summarisation with CI/CD | Python, Docker, GitHub Actions | Containerised pipeline with automated deployment via GitHub Actions |
| 🧠 AI Context Engine | Prevent AI conversation context from being lost between sessions | FastAPI, Chrome Extension, WebSockets | Full-stack browser tool that captures and restores AI session history across tabs |
Business Problem: How do you turn processed data into self-service tools that non-technical stakeholders can act on?
Interactive dashboards built for HR, job markets, and quick-commerce intelligence.
| Project | Problem Solved | Key Metrics Tracked | Links |
|---|---|---|---|
| 👥 HR Analytics Dashboard | Give HR teams a single view of workforce health across 8,950 employees | Headcount, attrition rate, gender distribution, department performance, education-pay correlation | Dashboard |
| 🛒 Zepto vs Blinkit | Help consumers and analysts compare platform pricing and delivery performance | Category dominance, discount patterns, pricing elasticity, delivery speed | Dashboard · BI Presentation |
| 💼 Job Market Analytics | Track which skills, roles, and geographies are growing or declining in demand | Skill demand trends, salary comparisons, geographic breakdown, YoY growth | Coming Soon |
I use Cursor and Claude Code as core development tools — enabling faster prototyping, automated code review, and AI-assisted debugging. I also use n8n for workflow automation and Google Sheets for quick ad-hoc analytics and stakeholder-ready reporting.
