Data Science with Gen AI in Mumbai

• What is Data Analytics vs Data Science vs Business Intelligence
• Roles and responsibilities: Business Analyst, Data Analyst, BI Developer
• Analytics workflow: Data collection → Analysis → Visualization → Insights → Action
• Real-world case studies from industry (finance, retail, e-commerce, healthcare)
• Career paths and salary benchmarks in analytics

• Installing Python, Jupyter Notebook, VS Code
• Package managers: pip, conda
• Setting up virtual environments
• Introduction to IDE best practices

• Language fundamentals: variables, data types, operators
• Control flow: if-else, loops (for, while)
• String manipulation and formatting
• Lists, tuples, dictionaries for data handling
• Functions: defining, parameters, return values

• NumPy arrays vs Python lists: performance benefits
• Array creation, indexing, slicing
• Mathematical operations on arrays
• Broadcasting concepts
• Statistical functions (mean, median, std, percentile)
• Practical: Working with multi-dimensional datasets

• Series and DataFrame fundamentals
• Creating DataFrames from CSV, Excel, SQL databases
• Indexing and selecting data
• Viewing and inspecting data (head, tail, info, describe)

• Handling missing values (dropna, fillna, interpolation)
• Removing duplicates
• Data type conversions
• Handling outliers

• Filtering and sorting
• Groupby operations and aggregations
• Pivot tables and crosstabs
• Merging, joining, concatenating DataFrames

• DateTime handling
• Resampling and rolling windows
• Trend analysis

• Understanding data distribution
• Identifying patterns and relationships
• Statistical thinking for analysts
• Types of analysis: univariate, bivariate, multivariate
• Hypothesis formation from data exploration

• Figure and axis objects
• Line plots, bar charts, histograms, scatter plots
• Customization: colors, markers, line styles
• Legend, labels, titles, annotations
• Grid and axis formatting
• Multiple plots: subplots and figure layouts

• Statistical visualization concepts
• Categorical plots: bar, box, violin, strip plots
• Distribution plots: histograms, KDE plots
• Relationship plots: scatter, regression plots, heatmaps
• Color palettes and styling
• FacetGrid for multi-dimensional analysis

• Plotly basics: interactive charts and hover information
• Creating dashboards with Streamlit
• Deploying Streamlit apps to cloud (Streamlit Cloud, Heroku)
• Real-time data updates
• Dashboard best practices

• Principles of data visualization
• Choosing the right chart for the message
• Narrative structure: context, conflict, resolution
• Presenting insights to non-technical stakeholders
• Creating compelling business reports

• Central tendency: mean, median, mode
• Spread: range, variance, standard deviation, IQR
• Skewness and kurtosis
• Outlier detection and handling (IQR method, Z-score)
• Weighted statistics

• Fundamentals: sample space, events, probability rules
• Conditional probability and Bayes' theorem
• Independence and dependence
• Real-world application scenarios

• Sampling concepts and distributions
• Confidence intervals: for mean and proportion
• Hypothesis testing: null vs alternative hypotheses
• T-tests, Z-tests, chi-square tests
• P-values and significance levels
• Type I and Type II errors

• SELECT, WHERE, ORDER BY, LIMIT
• Filtering with AND, OR, NOT, IN, BETWEEN
• DISTINCT and COUNT

• GROUP BY and aggregation functions (SUM, AVG, MAX, MIN, COUNT)
• HAVING clause
• INNER JOIN, LEFT JOIN, RIGHT JOIN, FULL OUTER JOIN
• Self-joins and multiple joins

• What is Generative AI vs Traditional ML
• GenAI applications across industries (content, code, analytics, customer service)
• The GenAI landscape: LLMs, Diffusion models, Multimodal models
• Business use cases and ROI analysis
• Ethical considerations and responsible AI

• History and Evolution
• From word embeddings (Word2Vec) to transformers
• GPT-3 vs GPT-4 vs Claude vs Llama evolution
• Understanding the transformer architecture (conceptual)

• Tokens and tokenization
• Attention mechanisms (high-level)
• Training process: next token prediction
• Context windows and token limits
• Temperature, top-k, top-p sampling explained

• *Same as Data Analytics Module 1-2, abbreviated*
• Python fundamentals for LLM applications
• NumPy for numerical operations
• Pandas for handling structured data
• Statistics essentials for model evaluation
• Setting up ML development environment (Jupyter, VS Code, Colab)

• Cloud-Based LLM APIs
• OpenAI API (GPT-4, GPT-3.5): authentication, rate limiting, cost optimization
• Google Cloud Vertex AI: Gemini models, deployment options
• Anthropic Claude API: working with long context windows
• Azure OpenAI: enterprise deployment

• Hugging Face ecosystem: model hub, model cards, licensing
• Llama 2, Mistral, CodeLlama: deployment and usage
• Running models locally: HuggingFace Transformers, Ollama
• Quantization: reducing model size for edge deployment

• Prompt Structure
• System prompts vs user prompts
• Role-based prompting
• Few-shot vs zero-shot prompting
• Chain-of-thought prompting for complex reasoning

• Prompt injection and security
• Template-based prompts for scalability
• Evaluating prompt quality
• Iterative prompt optimization

• Core Concepts
• LLMs, Chains, Agents, Memory
• Prompt templates and formatting
• Output parsing and validation
• Sequential chains and branching logic

• Integrating with OpenAI, HuggingFace, local models
• Building chatbots with conversation memory
• Document loaders and text splitters
• Chaining multiple LLM calls

• Why RAG?
• Hallucination problem in LLMs
• Keeping models current without retraining
• Domain-specific knowledge integration
• Cost optimization vs accuracy tradeoff

• Data ingestion pipeline
• Chunking and preprocessing strategies
• Embedding generation and semantic search
• Retrieval ranking and reranking
• LLM generation with context

• Introduction to Vector Databases
• Why embeddings and vector similarity matter
• Similarity metrics: cosine, L2, dot product
• Vector database landscape: Pinecone, Weaviate, Qdrant, Milvus, Chroma

• Creating and managing vector indexes
• Embedding models: OpenAI embeddings, open-source models
• Storing and retrieving vectors
• Metadata filtering and hybrid search
• Scaling vector databases

• End-to-End RAG Pipeline
• Document ingestion: PDF, web, database sources
• Text preprocessing: cleaning, chunking, overlap strategies
• Embedding storage and indexing
• Query expansion and multi-step retrieval
• Context ranking and relevance filtering
  

• Retrievers in LangChain
• Prompt engineering for RAG
• Answer generation with source attribution
• Handling multi-document retrieval

• Retrieval quality metrics: precision, recall, MRR, NDCG
• Generation quality: semantic similarity, BLEU, ROUGE, LLM-as-judge
• Latency and cost optimization
• A/B testing retrieval strategies
• 
  

• When to Fine-Tune?
• vs few-shot prompting vs RAG tradeoff analysis
• Cost-benefit analysis
• Use cases where fine-tuning excels

• Full fine-tuning: data requirements, compute resources
• Parameter-Efficient Fine-Tuning (PEFT):
• LoRA (Low-Rank Adaptation)
• QLoRA (Quantized LoRA)
• Prefix tuning, Prompt tuning
• Instruction tuning vs domain adaptation

• Task-specific metrics: classification, generation, QA
• Human evaluation frameworks
• Automated metrics: BLEU, ROUGE, METEOR, semantic similarity
• LLM-as-Judge evaluation
• Cost-quality tradeoff analysis

• Building chatbots with LangChain
• Memory management: conversation history, summaries
• Handling context limits
• Streaming responses
• Error handling and fallback strategies

• Application code (GitHub repository)
• Fine-tuned model or RAG pipeline
• API documentation
• Performance benchmarks and cost analysis
• Deployment guide (Docker + cloud platform)
• Presentation with business metrics

• Languages: Python 3.9+
• LLM Frameworks: LangChain, LangGraph, HuggingFace Transformers
• APIs: OpenAI, Google Vertex AI, Anthropic Claude, HuggingFace
• Vector Databases: Pinecone, Weaviate, Qdrant, Chroma
• Deployment: FastAPI, Docker, AWS/GCP/Azure
• Notebooks: Jupyter, Google Colab
• Monitoring: Langfuse, Arize

1. Understand LLM architecture and capabilities/limitations

2. Prompt engineer effectively for production applications

3. Build RAG systems with vector databases

4. Fine-tune models for specific domains

5. Evaluate GenAI application quality

6. Deploy and scale LLM applications

7. Design cost-effective AI solutions

8. Build production-ready chatbots and content generation systems

• Weekly coding assignments
• 3 mini-projects (RAG, Fine-tuning, Application)
• Capstone project with business analysis
• Portfolio-ready projects for LinkedIn/GitHub

• Solid Python programming skills
• Familiarity with ML/Data Science concepts (not required but helpful)
• Understanding of basic API usage
• No prior GenAI experience required
• Displaying DS_GenAI_Syllabus.md.