Advanced SQL — Window Functions + CTEs

ROW_NUMBER, RANK, DENSE_RANK, LAG/LEAD, running totals, moving averages, PARTITION BY. Common Table Expressions and query structuring. Cohort and retention queries. These separate ₹3 LPA analyst offers from ₹5 LPA ones — drilled through 25 interview-grade problems.

EDA — Exploratory Data Analysis Method

The professional EDA workflow: profile → clean → univariate → bivariate → outliers → correlations → hypotheses → summary. Feature understanding, data dictionaries, documenting assumptions. Two full EDA case studies from raw file to written findings.

Tableau

Tableau Public: connecting data, dimensions vs measures, calculated fields, filters, parameters, dashboards, stories. Power BI vs Tableau positioning for interviews. Publish one dashboard to a public Tableau profile — a second portfolio link.

Predictive Analytics — scikit-learn Hands-On

Applying the ML concepts practically: train/test split, Linear Regression (predict booking revenue), Logistic Regression (predict cancellation), k-Means (customer segments), model evaluation (accuracy, confusion matrix, R²). Analyst-level ML: build, evaluate, explain — not research-level math.

AI Fundamentals — Simply Explained

What is AI, ML, Deep Learning, LLMs — the hierarchy in plain English. How a language model works: trained on text, predicts next word, scales to billions of parameters. Why it gives different answers each time (temperature). Rule-based systems (old IT software) vs AI-powered systems (modern IT software). What training data is and why it matters.

How AI Helps in IT Service Projects

Real scenarios students will face: (1) Client wants a chatbot on their business website — build with Python + Gemini API. (2) Client has a product catalog Excel — customers want to query it in plain English. (3) Auto-generate professional service descriptions for SME providers. (4) Support ticket auto-classification for an IT helpdesk. Analytics angle: the same pattern generates automatic insight summaries under dashboards.

Prompt Engineering — Practical

System prompt vs user prompt. Role prompting: "You are a customer support agent for XYZ Clinic." Few-shot: give 2–3 examples for consistent output. Context injection: pass document or data text, ask questions. Output format control: "Always respond in JSON." Students write and test 10 prompts — hands-on, not slides.

Gemini API — Calling AI from Python

API key from Google AI Studio (free, no credit card). Install google-generativeai. 10-line Python call. Full-stack students wrap it in a Django view; analytics students wrap the same call in their notebook and Streamlit app. Full chain: data summary → prompt → Gemini → insight text. Powers the AI features in the capstone.

Copilot + ChatGPT + Claude — Daily Workflow

GitHub Copilot: write comment → Tab to accept → always read before using. Best for repetitive Pandas cleaning code, chart code, SQL queries. ChatGPT debug template: "Error: [X]. Code: [paste]. Expected: [Y]. Fix?" Claude for review: "Review my analysis notebook — logic issues?" Woven into every afternoon session from week 13.

Supervised Learning

Definition: model learns from labeled data — input + correct output pairs. How it trains: adjust weights to minimize prediction error. Real examples students relate to: predict service booking price (regression), classify support ticket as urgent/not urgent (classification).

Key algorithms (conceptual): Linear Regression — predict a number · Logistic Regression — yes/no output · Decision Tree — if/else logic at scale · k-Nearest Neighbors — vote by similarity.

Use cases: Predict customer churn for a CRM client · Classify support tickets by priority · Price prediction for a service marketplace · Spam detection for email marketing tools.

Unsupervised Learning

Definition: model finds patterns in data with no labels. Groups similar data without being told what the groups are. Real example: cluster customers by booking behavior — without knowing the groups in advance.

Key concepts: k-Means Clustering — group by distance · How cluster count is chosen · Dimensionality reduction (PCA — awareness only) · Anomaly detection concept.

Use cases: Customer segmentation for marketing clients · Product grouping for e-commerce apps · Detect unusual login patterns in a web app · Group similar support tickets automatically.

Reinforcement Learning

Definition: agent learns by trial and error — takes actions, gets rewards or penalties, improves over time. Taught conceptually only — no code implementation needed for analyst roles. Students understand why it powers recommendation engines and game AI.

Key concepts: Agent, environment, action, reward · Explore vs exploit trade-off · How reward shaping works · Real-world: recommendation systems.

Why it matters: Powers "recommended services" features · Used in dynamic pricing engines · Basis of chatbot self-improvement · Interview awareness question — very common.

DFS — Depth First Search

How it works: explore as deep as possible down one path before backtracking. Uses a stack (or recursion). Traverses a tree or graph branch by branch. Good for: finding if a path exists, maze-type problems, traversing nested data structures.

What students learn: Stack-based traversal logic · Recursive DFS implementation in Python · Pre-order, in-order, post-order tree walk · Time complexity: O(V+E).

Use cases: Crawling nested categories in a dataset · Decision tree traversal in ML · Pathfinding in service area mapping · Dependency resolution.

BFS — Breadth First Search

How it works: explore all neighbors at the current depth before going deeper. Uses a queue. Finds the shortest path between two nodes. Good for: social network connections, level-order tree traversal, finding nearest locations in a map.

What students learn: Queue-based traversal logic · BFS implementation in Python · Level-order tree traversal · Shortest path concept.

Use cases: Find nearest service provider in a marketplace dataset · Social graph connections · BFS in AI state-space search · Web crawler page discovery logic.

Streamlit Fundamentals

Turn a notebook into a web app in pure Python: st.title, widgets (selectbox, slider, date picker), st.dataframe, charts, layout columns, caching. Taught directly as: "Your analysis is done. Now let anyone interact with it in a browser."

AI-Powered Analytics App

Build the Zuvio Insights Pro app: file upload → Pandas cleaning → interactive charts → filters → the Gemini call from AI4 generating plain-English insight summaries under each chart → cancellation-risk prediction from the X4 model. One app that shows the entire course.

Streamlit Community Cloud + Tableau Public + GitHub

Streamlit Community Cloud: connect GitHub → pick repo → live app URL in minutes, free. Tableau Public profile with a published dashboard. GitHub as the portfolio hub linking everything. Students get permanent public links for resume and LinkedIn.

2–3 Week Live Project — Ether Services

Real client brief: analyze a client's sales/booking data, build a monthly reporting dashboard, automate a recurring Excel report with Python, or add AI insight summaries to an existing report. Real GitHub commits. Supervised by Ether Services team. Ends with signed certificate + a specific story for interviews.

GitHub Copilot

VS Code extension. Reads code context → suggests next lines. Best for: repetitive Pandas cleaning code, chart boilerplate, SQL queries. Saves 30–40% writing time on repetitive code. Key rule: always read what Copilot writes before accepting.

ChatGPT

Debug template: "Error: [X]. Code: [paste]. Expected: [Y]. What is wrong?" Also: explain unfamiliar errors, draft README, generate realistic test data for practice.

Claude

Best for: "Review my analysis notebook — logic issues?" "Is my statistical interpretation correct?" More thorough for reasoning review. Also excellent for polishing the written insight report.

Google Gemini API — teach first

Free tier: 15 req/min, 1M tokens/day. API key from aistudio.google.com — no credit card. Called from Python with google-generativeai. Used for all AI features: insight summaries, plain-English data Q&A, report drafting.

OpenAI API — optional

GPT-4o-mini: ~$0.60 per 1M tokens. ~₹50 total for a student project. Appears in more JDs. Needs a credit card — the reason Gemini is taught first.