SB
Spandana Bala
AI/Data Scientist
Hyderabad · India
Now at · Microsoft
Class 014 · DATA SCIENCE & AI AGENTS · RESEARCH + AGENTIC AI
Data Science
+ AI
Agents
Master end-to-end Data Science with Agentic AI. Build Python and PostgreSQL data foundations, ship statistical and ML experiments with proper cross-validation, engineer Deep Learning and NLP models with interpretability, and deploy a Data Science AI Agent that runs autonomous EDA and hypothesis generation.
3mo
duration
50+
modules
4.7/5
class rating
100k+
enrolled
Where our Data Science alumni work
Microsoft
Amazon
Salesforce
ServiceNow
Deloitte
Infosys
Accenture
TCS
Wipro
Capgemini
Cognizant
HCL
MicrosoftAmazonSalesforceServiceNowDeloitteInfosysAccentureTCSWiproCapgeminiCognizantHCL
What you leave with
Four things every Data Scientist grad walks away with.
01
Agent-Ready data science skills
Statistical rigour plus the full ML/DL/NLP and GenAI stack — NumPy, Pandas, PyTorch, TensorFlow, LangGraph, Claude Agent SDK, MCP — production-grade, not notebook demos.
02
A shipped project
A production-deployed Data Science AI Agent built with LangGraph + Claude Agent SDK + MCP that runs autonomous EDA, statistical tests, and feature engineering with a public verification URL.
03
Verifiable credential
2026 Agent-Ready rubric mapped to DP-100, AWS ML Specialty, Google Pro ML Engineer, and TensorFlow Developer, graded 1–5, with a public verification URL recruiters can check in 30 seconds.
04
Direct placement pipeline
GitHub + LinkedIn portfolio rewrite, DS-tuned resume rebuild, and warm intros into our 1,000+ hiring partners actively staffing Data Scientist and ML/AI Research roles.
3 MONTHS · FOUR PHASES · ONE DATA SCIENCE AGENT
From “plots a scatter chart” to ships intelligent data agents..
Weeks 1–2 · Data Foundations
Python + PostgreSQL for Data Scientists
- Python fundamentals, OOP, decorators, generators, and context managers
- Data structures and file handling for messy real-world CSV/JSON
- PostgreSQL — DDL, DML, JOINs, window functions, CTEs, PL/pgSQL
- Database design and query optimisation with EXPLAIN ANALYZE
YOU SHIPA production-quality Python codebase plus a normalised PostgreSQL schema with analytics queries — the foundation every later ML run sits on.
Weeks 3–7 · Scientific Toolkit
Power BI + Math/Stats + Python Libraries
- Power BI with DAX, time intelligence, and Row-Level Security
- Linear algebra, calculus, probability, and Bayes' theorem foundations
- Hypothesis testing, A/B tests, power analysis, and Bayesian inference
- NumPy, Pandas, Matplotlib, Seaborn, and Plotly for data work
YOU SHIPA rigorous EDA portfolio with hypothesis tests, A/B analyses with power calculations, and a Power BI dashboard suite.
Weeks 8–12 · ML + Deep Learning + NLP
Machine Learning + Deep Learning + NLP
- Regression, classification, and tree ensembles including XGBoost
- Unsupervised learning, clustering, PCA, t-SNE, and UMAP
- Model evaluation, Optuna tuning, MLflow tracking, SHAP and LIME
- Deep learning with CNNs, Transformers, and modern NLP pipelines
YOU SHIPA complete ML/DL/NLP portfolio — cross-validated models, a CNN image classifier, and a Transformer NLP model with attention.
Weeks 12–14 · GenAI + Agentic AI
Master the 2026 GenAI + Agentic AI stack — and ship a Data Science AI Agent that runs autonomous EDA, proposes experiments, and explains results in plain English.
Engineer with LLM APIs from OpenAI, Anthropic, Google GenAI, and DeepSeek. Master prompt engineering (zero-shot, few-shot, CoT, ReAct) and context engineering — the 2026 frontier discipline. Build production RAG pipelines with ChromaDB, Pinecone, Qdrant, and pgvector over your datasets, notebooks, and research papers. Master the 2026 production agent stack — LangGraph 1.0 (#1 production default), Claude Agent SDK (#2 MCP-native), CrewAI (#3 multi-agent crews). Wire it all through the Model Context Protocol (MCP) — 200+ server implementations, 97M+ monthly SDK downloads. Final project — a deployed Data Science AI Agent with MCP servers exposing your data sources, trained models, and statistical libraries. The data scientist’s force multiplier.
Partner orgs
(2026)54
Data Science
projects deployed300+
→ Placement
offers86%
Course curriculum
Seven sections. 65+ modules. The AI-native data science stack.
01
Python for AI & Data
The dominant language for data science. Master Python syntax, data structures,
and advanced programming concepts essential for AI and data science applications. 10 modules from
environment setup through advanced OOP — the language fluency that makes every later section
possible.
Python interpreter installation for Windows and Mac
Visual Studio Code + Jupyter for data science workflows
Python's 35 essential keywords
Identifiers and naming conventions
Variables and memory management fundamentals
Simple and complex data types
Type conversion and casting
Arithmetic, comparison, and logical operators
User input with input() function
Conditional statements — if, elif, else, match-case
while and for loops with range() function
break, continue, and pass statements
String definition and syntax rules
Positive and negative indexing
Slicing with start:end:step notation
Concatenation and repetition operations
f-strings and .format() for analysis output formatting
Immutability concept and implications
String methods — case conversion, search, checking, trimming, replacement,
split/join
Critical for text data preprocessing in NLP analyses
Lists — creation, indexing, slicing, modification operations
List comprehensions for elegant data transformation
Sorting, reversing, copying patterns
Tuples — creation and basic operations
Tuple packing and unpacking
Performance advantages over lists
Use cases for immutability in scientific computing
Dictionaries — creation, access, operations
Dictionary comprehensions
Nested dictionaries for structured experimental data
Essential for representing experimental configurations and results
Sets and the UUU properties (Unique, Unordered, Unindexed)
Mathematical operations — union, intersection, difference
Subset and superset checks
Applications in deduplication and set algebra
Collections module — namedtuple, Counter, defaultdict, deque
Iteration protocol and custom iterators
Generators using yield for memory-efficient processing of large datasets
Generator expressions and pipelines
Lambda functions for anonymous operations
Higher-order functions — map(), filter(), reduce()
Functional patterns for reproducible data transformations
Function definition, parameters, return values
Default arguments, *args, **kwargs
Variable scope (LEGB rule)
First-class functions and higher-order patterns
Recursion and recursive design patterns
Type hints (Python 3.5+) — essential for self-documenting analysis code
Documenting functions with docstrings (PEP 257)
Built-in modules, user-defined modules, packages
pip for package management
requirements.txt for reproducible analyses
Virtual environments for project isolation
Reproducibility is non-negotiable in data science — invest here
CRUD operations with open()
File modes and pathlib
CSV files — csv module for tabular data ingestion
JSON files — json module for API responses and config files
The most common data input formats for data science work
Exception Handling — robust error handling for unreliable data sources
Decorators — for logging experiment runs, timing functions, caching results
Generators Deep Dive — memory efficiency for processing larger-than-RAM datasets
Context Managers — proper resource management for database connections, file
handles
These four patterns are what every senior data science role expects fluency in
Object-Oriented Programming models real-world entities through classes and
objects — essential for structuring data pipelines, custom estimators, and reproducible analysis
workflows
Classes, objects, methods, special methods
Instance variables vs class variables
Encapsulation — access modifiers control data visibility
Inheritance — single, multi-level, and multiple inheritance
Abstraction — abstract classes and methods
Polymorphism — method overriding and duck typing
scikit-learn's estimator pattern relies on OOP — master it before Section 6
02
SQL for AI & Data
The data backbone of every data scientist's daily workflow. Five modules
covering PostgreSQL from foundations through advanced analytics queries — window functions, CTEs,
programming with PL/pgSQL — the data layer that powers your experiments and your A/B test analyses.
Databases, DBMS, RDBMS — concepts and terminology
ACID properties — Atomicity, Consistency, Isolation, Durability
PostgreSQL setup, psql, pgAdmin 4, DBeaver
Data types — numeric, character, date/time, boolean, JSON, arrays
Constraints — PRIMARY KEY, FOREIGN KEY, UNIQUE, NOT NULL, CHECK, DEFAULT
SELECT statements and column projection
WHERE clauses with operators and conditions
Built-in functions — string, numeric, date, conditional
Aggregates — COUNT, SUM, AVG, MIN, MAX
GROUP BY and HAVING for aggregation
Window functions — ROW_NUMBER, RANK, DENSE_RANK, LAG, LEAD — essential for cohort
analysis and time-series work
JOIN operations — INNER, LEFT, RIGHT, FULL OUTER, CROSS, SELF
Subqueries — scalar, row, table subqueries
CTEs (Common Table Expressions) — the data scientist's favourite for readable
analytical queries
Recursive CTEs for hierarchical data
Set operators — UNION, UNION ALL, INTERSECT, EXCEPT
DML — INSERT, UPDATE, DELETE patterns
Transactions — atomicity for data integrity
ALTER TABLE for schema evolution
Indexes — B-tree, Hash, GiST, GIN
Views — virtual tables, materialized views for caching analytical aggregates
PL/pgSQL — variables, control structures, exception handling
Stored procedures vs functions
Triggers — automation for data quality checks
ER modelling, normalization (1NF, 2NF, 3NF)
OLTP vs analytics workload patterns
Star schema design for analytics
Query plan analysis with EXPLAIN and EXPLAIN ANALYZE
Index strategies
VACUUM, ANALYZE, partitioning
A scientist who can't read EXPLAIN ANALYZE is at the mercy of slow queries
forever — invest here
03
Power BI for Data Analysis
The data scientist's tool for stakeholder communication. Power BI provides a
comprehensive platform for connecting to data sources, creating interactive visualisations, and
sharing insights across organisations through intuitive dashboards and reports — turning your
analysis into the kind of executive dashboards that drive business decisions.
BI fundamentals and modern analytics approaches
Power BI components — Desktop, Service, Mobile, Gateway
Interface navigation, workspace setup, first report creation
Desktop versus Service capabilities
File, database, cloud, web sources
Connection modes — Import, DirectQuery, Live Connection
Performance considerations for analytical workloads
M language fundamentals
Step-by-step transformations
Data profiling and quality assessment
Reshaping operations
Star schema design for analytical reporting
Relationships — one-to-many, many-to-many
Hierarchies — date, geography, organisational
Calculated columns vs measures
Core visualisations — bar, line, scatter, pie, treemap, waterfall
Choosing the right chart for the right insight
Interactive elements — slicers, filters, bookmarks, drill-through
DAX syntax and structure
Aggregation, logical, text, date/time functions
CALCULATE and FILTER — the heart of DAX
Variables for performance and readability
Year-over-year, quarter-over-quarter comparisons
Period-over-period growth calculations
Custom calendar handling
Iterator functions (SUMX, AVERAGEX, COUNTX)
ALL, ALLEXCEPT for filter manipulation
AI visuals — Q&A, Key Influencers, Decomposition Tree
Workspaces, apps for distribution
Sharing and access management
Row-Level Security (RLS) — dynamic security with USERPRINCIPALNAME()
Sensitivity labels
DAX performance tuning, composite models
Foundation for the Microsoft Certified: Power BI Data Analyst Associate (PL-300)
certification
04
Math & Stats for AI & Data
The mathematical foundation that distinguishes Data Scientists from data
analysts. Four modules covering linear algebra, calculus, probability theory, and inferential
statistics — including Bayesian methods — the rigorous foundation behind every meaningful analytical
recommendation.
Set theory and logical operations
Functions and graphs — linear, polynomial, exponential, logarithmic
Understanding mathematical relationships
Vector operations, dot product, magnitude
Matrix creation, operations, transpose, inverse, determinant
Matrix algebra and applications in PCA, regression
Systems of linear equations
Gaussian elimination method
Eigenvalues and eigenvectors
Diagonalisation and matrix decomposition
Introduction to PCA intuition
Limits, derivatives, partial derivatives
The chain rule — the mathematics of model training
Optimisation foundations
Definite and indefinite integrals
Probability density functions
Critical points and extrema
Gradient descent — the engine of ML training
Higher-dimensional optimisation
Sample spaces, events, probability axioms
Conditional probability and Bayes' Theorem
Independent vs dependent events
The probability chain rule
Discrete distributions — Bernoulli, Binomial, Poisson
Continuous distributions — Normal, Uniform, Exponential
Distribution properties — mean, variance, skewness
Central Limit Theorem
Joint and marginal distributions
Covariance and correlation
Multivariate Normal distribution
Measures of central tendency, spread
Distribution shape — skewness, kurtosis
The most important module for data scientists — get this right and your analyses
are trustworthy
Null and alternative hypotheses
Type I and Type II errors
p-values, significance levels, statistical power
t-tests (one-sample, two-sample, paired)
ANOVA for comparing multiple groups
Chi-square tests for categorical data
Non-parametric tests — Mann-Whitney, Kruskal-Wallis
A/B Testing — proper experimental design
Power analysis and sample size calculation
Multiple testing correction (Bonferroni, FDR)
Sequential testing pitfalls
p-hacking and reproducibility
Confidence intervals — construction and interpretation
Bootstrap confidence intervals
Bayesian vs frequentist statistics
Priors, likelihoods, and posteriors
Bayesian A/B testing — modern best practice
Introduction to MCMC concepts
PyMC and Stan ecosystems
The hypothesis testing and Bayesian methods you learn here are what make every
later ML evaluation defensible to scientific scrutiny
05
Python Libraries for AI & Data
The Python ecosystem that powers daily data science work. Five modules covering
NumPy for vectorised computing, Pandas for data manipulation, and the visualisation trio
(Matplotlib, Seaborn, Plotly) — your daily-driver libraries.
N-dimensional arrays — the foundation of all numerical Python
Array creation, indexing, slicing
Vectorised operations and broadcasting
Aggregations along axes
Linear algebra with np.linalg
Why vectorisation is 100x faster than Python loops
Series and DataFrame structures
Index manipulation and MultiIndex
Data loading — read_csv, read_excel, read_sql, read_json
head, tail, info, describe
Handling missing data — multiple imputation strategies
Removing duplicates, type conversion, string operations
groupby operations — the split-apply-combine pattern
Pivot tables and crosstabs
Merging and joining DataFrames
Reshaping — pivot, melt, stack, unstack
Time series — DateTime indexing, resampling, rolling windows
Time zone handling
Figure and Axes architecture
Object-oriented vs pyplot interfaces
Plot types — line, scatter, bar, histogram, box, heatmap
Customisation, annotations, legends
Subplots with subplots and GridSpec
Saving figures at publication quality
Distribution plots — histplot, kdeplot, boxplot, violinplot
Relational plots — scatterplot, lineplot, relplot
Categorical plots — barplot, countplot, pointplot
Matrix plots — heatmap, clustermap
FacetGrid for multi-panel comparative plots
PairPlot and JointPlot for distribution comparisons
Plotly Express for quick interactive plots
Plotly Graph Objects for custom control
Subplots and dashboards
Animation frames for time-series visualisations
Plotly Dash for web-deployed analytical dashboards
06
Machine Learning
The heart of data science — 10 modules covering supervised, unsupervised, and
reinforcement learning, with proper cross-validation, hyperparameter tuning, and model
interpretability (SHAP, LIME). Every module emphasises scientific rigour, reproducibility, and
evidence-based modelling decisions.
What is Machine Learning? AI vs ML vs Deep Learning
The scientific method applied to learning algorithms
The ML workflow — data → features → model → evaluation → interpretation
Supervised learning with labelled data
Unsupervised learning for pattern discovery
Reinforcement learning through interaction
Models, parameters, and hyperparameters
Role of optimisers in training
Train/validation/test splits — proper experimental design
Handling missing values — multiple imputation strategies
Encoding categorical variables — one-hot, label, target encoding
Feature scaling — StandardScaler, MinMaxScaler, RobustScaler
Feature engineering — polynomial features, binning, datetime features
Class imbalance — SMOTE, undersampling, class weights, threshold tuning
Simple and multiple linear regression
Ordinary Least Squares (OLS) estimation
Coefficient interpretation
Assumption checking — linearity, normality, homoscedasticity, independence
Diagnostic plots — residual plots, Q-Q plots
Train/test split and cross-validation
Regression metrics — MAE, MSE, RMSE, R², MAPE
Bias-variance tradeoff
Learning curves for model diagnosis
Gradient descent — batch, stochastic, mini-batch
Learning rate selection and convergence diagnostics
Multivariate regression with multicollinearity (VIF)
Interaction terms and their interpretation
Polynomial regression and feature transformation
Bias-Variance Tradeoff fundamentals
Understanding overfitting vs underfitting
Ridge Regression (L2) and Lasso Regression (L1)
Elastic Net combining both
Cross-validation for hyperparameter selection
GridSearchCV, RandomizedSearchCV, Optuna for Bayesian optimisation
Binary vs multi-class classification
Logistic regression and sigmoid function
Log Loss (Binary Cross-Entropy)
Decision boundary visualisation
Multi-class strategies (One-vs-Rest, One-vs-One, Softmax)
Naive Bayes — probabilistic classification based on Bayes' Theorem
Types of Naive Bayes (Gaussian, Multinomial, Bernoulli)
Laplace smoothing for zero probabilities
Applications in spam detection and sentiment analysis
Tree construction algorithms (ID3, C4.5, CART)
Splitting criteria — Gini impurity, entropy, information gain
Tree pruning to prevent overfitting
Interpretability advantages — trees you can explain to executives
Random Forests — ensemble methods, bagging and bootstrap aggregation
Feature importance ranking
Out-of-bag (OOB) evaluation
Gradient Boosting — XGBoost, LightGBM, CatBoost — the production champions
Boosting vs bagging
Early stopping and learning rate scheduling
Feature importance in boosted models
Linear and non-linear SVMs
Kernel trick — RBF, polynomial, sigmoid kernels
Soft-margin classification
K-Nearest Neighbours — distance metrics and weighting
Curse of dimensionality
Optimal k selection
Classification metrics — accuracy, precision, recall, F1-score
ROC curve and AUC
Precision-recall curves for imbalanced data
Confusion matrix interpretation
Calibration curves and threshold tuning
SHAP (SHapley Additive exPlanations) — game-theoretic feature attribution
LIME (Local Interpretable Model-agnostic Explanations)
Partial dependence plots
Permutation importance
Interpretable models build stakeholder trust — and meet regulatory requirements
(EU AI Act)
Discover hidden patterns in unlabelled data through clustering and dimensionality
reduction
K-Means Clustering and elbow method
Hierarchical Clustering and dendrograms
DBSCAN (density-based clustering)
Gaussian Mixture Models (GMM)
Clustering evaluation metrics — silhouette, Davies-Bouldin
Principal Component Analysis (PCA) for feature extraction
t-SNE for visualisation
UMAP for non-linear manifold learning
Curse of dimensionality
Feature extraction vs feature selection
RL fundamentals and paradigm
Agent, Environment, State, Action, Reward
Markov Decision Process (MDP)
Q-Learning Algorithm and Bellman equation
Exploration vs Exploitation tradeoff
Model persistence (pickle, joblib, ONNX)
Creating prediction APIs (Flask/FastAPI)
Model versioning and A/B testing
Model monitoring and drift detection
ML best practices and ethics
Responsible AI principles
MLflow for reproducible experiment tracking
Section Project — build, evaluate, and deploy a complete ML model with SHAP
interpretability, proper cross-validation, and a publication-quality results report
07
Deep Learning & NLP
Where data science meets the frontier. 10 modules covering neural networks from
scratch through PyTorch/TensorFlow, CNNs for computer vision, RNNs and Transformers for sequence
work, and complete NLP pipelines including named entity recognition — all with the interpretability
and rigour data scientists demand.
Introduction to artificial neural networks and perceptrons
Activation functions — sigmoid, tanh, ReLU, and variants
Forward propagation and backpropagation algorithms
Gradient descent and optimisation techniques (SGD, Adam, RMSprop)
Loss functions and performance metrics
Building ANNs from scratch in NumPy — understand before you use
Overfitting, underfitting, and regularisation
Introduction to PyTorch and TensorFlow
Tensors, computational graphs, and automatic differentiation
Building neural networks with PyTorch nn.Module
Data loading, batching, and augmentation
Training loops, validation, and checkpointing
GPU acceleration and performance optimisation
Debugging neural networks
Image representations and pixel manipulation
Image augmentation techniques
Transfer learning fundamentals
CNN architecture — convolution, pooling, padding, stride
Famous architectures — LeNet, AlexNet, VGG, ResNet, Inception
Transfer learning with pre-trained models
Image classification end-to-end
Attention visualisation and feature maps for interpretability
Object detection introduction (YOLO, Faster R-CNN)
RNN architecture and challenges
Vanishing/exploding gradients
LSTM (Long Short-Term Memory) networks
GRU (Gated Recurrent Units)
Bidirectional RNNs
Time series prediction with RNNs
Attention mechanism fundamentals
Self-attention and multi-head attention
Transformer architecture — encoder, decoder, positional encoding
Why Transformers replaced RNNs for most NLP
Introduction to BERT and GPT architectures
Text preprocessing and tokenisation
Bag of Words and TF-IDF
Word embeddings — Word2Vec, GloVe, FastText
Training custom word embeddings
Language modelling fundamentals
N-gram models and neural language models
Evaluation metrics for NLP
Text classification pipeline design
Feature extraction for text data
CNN for text classification
RNN and LSTM for text classification
Sentiment analysis techniques
Multi-class and multi-label classification
Handling imbalanced text datasets
Model evaluation and error analysis
Sequence-to-sequence architecture deep dive
Machine translation fundamentals
Encoder-decoder models
Handling long sequences
Beam search and decoding strategies
Evaluation metrics (BLEU, ROUGE)
Training strategies
Named Entity Recognition (NER) fundamentals
Part-of-speech tagging
Sequence labelling with RNNs and LSTMs
BiLSTM-CRF for NER
Information extraction
Relation extraction
Domain-specific NER systems
Evaluation metrics
Section Project — a computer vision model (CNN) plus an NLP model (BERT-based),
both deployed with proper evaluation, attention visualisation, and a publication-quality results
report
08
Generative AI & Agentic AI
The frontier — and the culmination of the data science programme. 10 modules
covering the complete 2026 GenAI engineering stack tuned for data science work: frontier models,
prompt engineering, RAG over scientific literature, agent frameworks, and the Model Context
Protocol. The named Data Science AI Agent project lives here.
Narrow AI — image classifiers, speech recognition (pre-2022)
Generative AI — LLMs, image/video/audio generation (post-2022)
Agentic AI — Plan / Reason / Act / Learn loops (post-2024)
2022 inflection point — ChatGPT launch
2024 inflection point — Agentic emergence
For Data Scientists — AI assistants that draft EDA reports
Hypothesis generation from data summaries
Autonomous experiment design (with human approval)
GPT-5.5 — Terminal-Bench 2.0 leader at 82.7%. Best for autonomous research agents
Claude Opus 4.7 — SWE-bench Pro leader at 64.3%, lowest hallucination rate at
36%. Best for long-form analysis writing and statistical reasoning
Gemini 3.1 Pro — 2M+ token context window. Best for ingesting entire research
papers and codebases
Open-source frontier — Llama 4, DeepSeek, Mistral, Qwen — for sensitive data
workflows
Intelligent routing — Opus 4.7 for analytical writing and interpretation
GPT-5.5 for autonomous research agents
Gemini 3.1 Pro for massive-context analysis
Perplexity — citation-grounded research
NotebookLM — long-document analysis for papers
Eight lessons culminating in a portfolio-ready 30+ prompt library for data
science work
Fundamentals — Context + Task + Examples + Format + Constraints
Core Techniques — Zero-shot, few-shot, CoT, ReAct, Tree-of-Thought
System Prompts — persistent persona design, guardrails
Multimodal — reading plots, dashboards, hand-drawn diagrams
Hallucination & Context — grounding for accurate analyses
Domain & Library — data science prompt patterns
Context Engineering — the 2026 frontier discipline; managing what enters the
LLM's context window for accurate scientific reasoning
ChatGPT, Claude, Gemini for daily data science work
AI for analytical writing — reports, summaries, executive memos
Research with Perplexity for literature reviews
AI for code — GitHub Copilot, Cursor for analysis acceleration
Building data-science-specific AI workflows
Reading plots and dashboards with vision models
Analysing whiteboard sketches and hand-drawn DAGs
OCR for legacy research documents
Image generation for paper figures
Audio — Whisper for interview transcription
Multimodal LLMs now read your scatter plots and identify outliers better than
novice analysts
Hallucination — when an LLM invents a statistical fact
Prompt injection through documents
Privacy — keeping research data out of public LLMs
Regulatory landscape — EU AI Act, India DPDP Act
Validating AI-generated analyses — the data scientist's new responsibility
Streamlit — rapid prototyping for internal data science apps
FastAPI — production-grade Python API for AI services
Building chatbots for analytical Q&A
Building diagnostic agents for data quality
Build and deploy a Streamlit + FastAPI internal tool
LLM APIs in production — OpenAI, Anthropic, Google GenAI, DeepSeek Python SDKs
API patterns — completions, streaming, function calling, structured outputs
Rate limits, retries, exponential backoff
Function calling & structured outputs — the 2026 production pattern for reliable
JSON
Pydantic-validated structured outputs
Embeddings — OpenAI text-embedding-3-large, Voyage, Cohere
Vector databases — ChromaDB, Pinecone, Qdrant, pgvector
HNSW, IVF indexing strategies
RAG pipeline for data science — canonical flow applied to research literature,
dataset documentation, and analytical reports
Chunking strategies — fixed-size, semantic, hierarchical
Hybrid search (BM25 + embeddings)
Re-ranking with cross-encoders
Agentic RAG — self-improving retrieval over scientific literature
Multi-step retrieval — find relevant papers, extract methods, compare findings
Project — Internal Data Science RAG App: RAG over your team's analyses,
datasets, and research papers
LangGraph 1.0 — complex stateful workflows — the production default for
autonomous data science
Claude Agent SDK — deepest MCP integration, extended thinking for complex
analysis
CrewAI — multi-agent crews; use case: a "data science team" of agents (EDA Lead,
Hypothesis Generator, Modeller, Reviewer)
Semantic Kernel / Microsoft Agent Framework
Pydantic AI — type-safe Python
ReAct — investigate a dataset, then propose an analysis
Plan-and-Execute — generate a multi-step analysis plan
Reflection loops — agent reviews its own analysis before finalising
Multi-agent collaboration — EDA agent proposes, Modeller agent builds, Reviewer
agent critiques
Human-in-the-loop checkpoints — humans approve every interpretation-critical step
MCP — open standard for connecting agents to tools, data, and systems
Proposed by Anthropic in late 2024
Stewarded by Linux Foundation — 200+ servers, 97M+ monthly SDK downloads
Build an MCP server exposing your trained models for agent inspection
Build an MCP server exposing your data sources and notebooks
Connect LangGraph agents to multiple MCP servers
Use Claude Agent SDK's deepest native MCP integration
DATA SCIENCE AI AGENT CAPSTONE — multi-agent Data Science AI Agent using
LangGraph + Claude Agent SDK with MCP servers exposing your data sources, trained models, and
analysis notebooks
The agent runs autonomous EDA, suggests appropriate statistical tests, proposes
feature engineering, and explains results in plain English — with human approval gates for every
interpretation-critical step
Frontend with Streamlit, backend with FastAPI, observability via LangSmith —
public verification URL with the 2026 Agent-Ready rubric — the named project for the entire Data
Science & AI programme
Tools you'll master
32+ ML, DL & AI tools, one production project.
Real-time projects
You don't watch videos. You ship software.
Three full-production projects, each threaded through the entire curriculum. By the project, you've built the whole stack around them.
Hero project · weeks 3–12
Production ML system: train → serve → monitor
Build an end-to-end production ML pipeline — reproducible training, model serving, drift monitoring, and an LLM augmentation layer that explains predictions and answers analyst questions.
01Reproducible training pipeline — DVC-versioned data & features, MLflow experiments, Optuna hyperparameter search, Weights & Biases tracking.
02Model-serving stack — Triton or BentoML server packaging your model + ONNX runtime, with FastAPI gateway, autoscaling on Kubernetes.
03Drift + quality monitoring — Evidently dashboards for data & concept drift, alerting hooks into Slack, retraining triggers.
04LLM augmentation layer — a small Hugging Face pipeline + a LangChain RAG layer that explains predictions and answers analyst questions.
Outcome: AUC +12 pts vs baseline
p95 inference: <120ms
Reviewer: ML Engineering panel
PyTorchMLflowTritonEvidentlyLangChain
Enterprise · weeks 6–11
NLP & LLM fine-tuning
Train a domain-tuned transformer end-to-end — Hugging Face PEFT/LoRA fine-tuning, evaluation on golden datasets, ONNX export, served via Triton with token-level latency dashboards.
Hugging FacePEFTONNXTriton
Real-time · weeks 8–12
Real-time recommendation system
Build a candidate-generation + re-ranking recommender on Spark + Feast feature store, served on SageMaker / BentoML, with online evaluation and Evidently drift monitoring.
SparkFeastBentoMLSageMaker
Project · weeks 11–12
Your ML system in a real partner org.
Pick a real partner ML problem. Deploy a production system end-to-end — feature store, training pipeline, model serving, drift monitoring, LLM explanation layer — into a partner team that's running it for real users.
Download the real world project
Full scope, sample partner orgs, weekly milestones, and grading rubric — PDF, 14 pages.
2026: 220+ deployed76% → placement offers
Your instructor
Taught by engineers who shipped agentic AI to production.
MK
Manikanta Kona
Founder, Digital Lync · AI & Data Science Architect
Python · PyTorch · TensorFlow · scikit-learn · Hugging Face · LangChain
"A 2026 data scientist doesn't stop at notebooks. They ship the training pipeline,
stand up the model behind FastAPI, monitor drift in production, and wire an LLM into the loop so the
business actually understands what the model is saying. That's the bar I teach to, every cohort."
15 yrs
AI & DATA SCIENCE
2,400+
LEARNERS
4.9 /5
RATING
Manikanta is the founder of Digital Lync and brings 15 years of applied AI & data science from AT&T, Salesforce, Cox Communications, and Broadcom — where he led recommendation, fraud, forecasting, NLP and computer-vision systems for Fortune-500 banks, telcos, and insurers. Most recently he architected production ML pipelines that pair classical and deep models with an LLM augmentation layer that explains predictions to business stakeholders.
His classes get you two things other programs don't give you: a founding architect who still ships production ML, and a curriculum rewritten every quarter to match what hiring managers actually ask about — credentials like AWS Machine Learning Specialty, Azure AI Engineer, Databricks ML Associate, TensorFlow Developer, and Pragmatic AI Engineer included. M.S. in Engineering, Purdue University.
RK
Ravi Krishna
Chief Technologist, Digital Lync · ML Engineering & MLOps Lead
PyTorch · MLflow · Triton · ONNX · Spark · MLOps · LLM Fine-tuning
"MLOps is where data science stops being a notebook and starts being a system —
reproducible training pipelines you can re-run a year later, model serving you can stake an SLA on, drift
monitoring that's quiet on purpose, and an LLM layer that explains decisions to the people who own the
business outcome. That's what I teach."
10 yrs
ML ENGINEERING
1,800+
LEARNERS
4.8 /5
RATING
Ravi is Chief Technologist at Digital Lync, where he leads the ML engineering and MLOps practice. After 8 years building and running production ML pipelines, he stepped into the Chief Technologist seat to wire MLflow, Triton, Evidently, and Hugging Face into the way ML teams actually work — feature stores that stay accurate through retrains, drift monitoring that filters noise before it hits humans, and serving stacks that on-call engineers don't fight with.
His MLOps modules are built from real production post-mortems, not slide decks. Expect to leave with working training pipelines, model serving on Triton/BentoML, drift dashboards in Evidently, and an LLM fine-tuning workflow you can stake an SLA on. Ten years at Digital Lync, eight of them shipping production ML — Hyderabad-based, hands-on, and known for the unglamorous parts of data science that everyone else skips.
HIRING PARTNERS · INDUSTRY VOICES
What ML & data science employers say about Digital Lync grads.
Real feedback from data and ML leaders at AI-first companies and the firms hiring our AI & Data Science graduates.
Digital Lync grads ramp 40% faster on production ML deploys than typical data science hires. Best AI & Data Science pipeline in India.
Aakash Mehta, Engineering Director, Microsoft
We've onboarded 80+ Digital Lync alumni in 18 months. Lowest ramp time we've seen for production ML systems and drift monitoring practices.
Anita Sharma, Senior Manager, Deloitte
The AI & Data Science programme is comprehensive — PyTorch, MLflow, Triton, MLOps. Grads come pre-trained for production ML & LLM engineering.
Rahul Bhatt, Solutions Lead, Mphasis
Their MLOps + drift monitoring track produces PMs who ship production-grade ML systems on day one. Rare combination of modeling rigor and engineering craft.
Deepak Pillai, Senior Architect, TCS
What sets Digital Lync apart is the production ML layer baked into the data science track. Our enterprise clients ask for exactly this profile.
Suresh Menon, Practice Lead, Accenture
Their AWS ML Specialty + Pragmatic AI Engineer prep is rigorous, and the shipped project — training pipeline, model serving, drift monitoring — is what closes interviews for us.
Vikram Iyer, Director, Infosys
Digital Lync's Data scientists ship production ML systems twice as fast in the first 90 days. Our internal modeling metrics back this up clearly.
Lakshmi Nair, VP Engineering, Wipro
Best AI & Data Science pipeline we've sourced from in India. Their projects are real production deploys, not notebooks.
Karthik Subramanian, Engineering Director, Cognizant
Strong PyTorch and MLOps foundation. Their AI/Data grads need almost zero ramp time on enterprise ML engagements with us.
Arun Joshi, Practice Director, Capgemini
We've placed 40+ Digital Lync alumni across our ML and watsonx engineering teams. Strong fundamentals, sharp on eval and drift monitoring.
Sanjay Verma, Talent Director, IBM
ML systems + drift monitoring is exactly the talent gap we've been struggling to close. Digital Lync is filling it for us reliably.
Anjali Desai, Practice Head, LTIMindtree
Their AI/Data Science track delivers engineers who navigate PyTorch, MLflow, and Triton on customer engagements unsupervised.
Ramesh Iyer, Senior Manager, Tech Mahindra
Hired 25+ Digital Lync graduates for our ML engineering practice. Strong on PyTorch, sharp on MLflow, fluent in MLOps.
Geetha Pillai, Talent Acquisition Lead, Cyient
Digital Lync grads who blend ML systems with Azure ML evals land production-ready on day one. Rare combination, well-trained.
Priya Reddy, Talent Lead, Microsoft
03Program certifications
An Agent‑Ready credential, not a participation trophy.
Digital Lync · Institute Certificate
Agent‑Ready AI & Data Scientist
Presented to
Spandana Bala
For the successful training, deployment, and monitoring of a production ML system —
reproducible training pipeline, model-serving stack, drift monitoring, and LLM augmentation — evaluated against the AWS ML Specialty, Azure AI Engineer, Databricks ML Associate, and Pragmatic AI Engineer credential rubrics.
Manikanta Kona
CEO · Digital Lync
AGENT
READY
2026
READY
2026
01
Industry‑recognized
Co‑branded with the ML engineering community and mapped to AWS ML Specialty
and Pragmatic AI Engineer credentials — names that hiring managers already scan for on resumes.
02
Project artifact included
Every certificate carries your shipped project — training pipeline, model serving, drift monitoring, LLM augmentation — with
a link to the live partner-org deployment. Proof, not a promise.
03
Enhanced skill validation
Graded against the 2026 Agent‑Ready rubric: training pipelines, model serving, drift monitoring, LLM augmentation,
MLOps automation. No pass/fail — a level 1‑5 band.
04
Verifiable on a public URL
Each credential has a public verification page recruiters can check in 10 seconds — no PDF
back‑and‑forth.
04Job placement support
Your first AI/Data Science offer isn't a lottery ticket. It's a built process.
GitHub, LinkedIn, resume — and most importantly, warm intros into AI labs and ML-heavy product orgs. Our
placement team works your search like an account, not a helpdesk.
01 / GITHUB & PORTFOLIO
A portfolio, not a graveyard.
Guidance on building a portfolio that showcases your training pipeline, model server, drift dashboard, LLM augmentation layer, and a public verification URL — reviewed 1:1, not via template.
02 / RESUME PREP
Rewrite, don't proofread.
A one-page resume rebuilt around the ML systems you shipped (training pipelines, model serving, drift monitoring), the partner-org project, and the business outcome. Reviewed by ML engineers who've read 10,000+ resumes.
03 / LINKEDIN + INTROS
Where most opportunities actually live.
Profile tuning plus direct warm introductions into AI labs and ML-heavy product orgs — Microsoft, Anthropic, OpenAI partners, Hugging Face, Databricks, Snowflake, Scale AI, NVIDIA, Stripe, Razorpay, plus services that staff data science teams (Deloitte, Accenture, Cognizant, TCS). You leave with recruiter contacts, not a generic "good luck."
AI & Data Science alumni
Hundreds of ML & data science careers launched — here are eight.
NV
Naveen Vedala
Senior ML Engineer
Hyderabad · India
Now at · Atlassian
TA
Tejashwini Addla
Staff Data Scientist (LLM)
Hyderabad · India
Now at · Salesforce
TD
Tharunesh Dillikar
Principal ML Engineer
Seattle · United States
Now at · Scale AI
MM
Mujahed Mohammed
NLP Engineer
Hyderabad · India
Now at · Databricks
BK
Bhargav Kumar Murala
MLOps Engineer
Hyderabad · India
Now at · Adobe
SL
Sai Manasa Leburi
Computer Vision Engineer
New York · United States
Now at · Hugging Face
RD
Rahul Dhamma
Recommendation Systems Engineer
Hyderabad · India
Now at · NVIDIA
Our locations
Come chat with us — over coffee, or over Zoom.
One flagship campus in Hyderabad, plus online Principal ML Engineer cohorts running on Indian and US timezones.
Flagship campus
Hyderabad
2nd Floor, Hitech City Road · Above Domino's · Opp. Cyber Towers, Jai Hind Enclave ·
Hyderabad, Telangana
Call
+91 90003 29956
US desk
+1 858 666 6719
Hours
Mon–Sat · 9am–9pm
Online class
Global
Weekend and evening AI/Data Science cohorts running on IST and PST. Every online cohort ships
the same shipped project — training pipeline, model serving, drift monitoring, LLM augmentation — as the on‑campus track.
Timezones
IST & PST
Format
Live + 1:1 mentorship
Next class
25 May 2026
FAQ
Questions we actually get — answered honestly.
Straight answers on prerequisites, the ML/DL stack, certifications, and placement. If something's missing, book a 20-minute advisor call — no slides, no pitch.
Do I need a CS or stats background?
No on both counts. Roughly 40% of every class comes from non-CS streams — engineering, math, physics, BCom, BBA, and self-taught modelers. Weeks 1–2 cover the NumPy/Pandas fundamentals, scikit-learn patterns, and the ML training loop from scratch. What you do need: consistency and 12–15 hours a week.
Will I actually ship production ML, or only do notebooks?
You actually ship to production. Every learner builds a reproducible training pipeline with DVC + MLflow + Optuna, packages the model with ONNX + Triton/BentoML, deploys it on Kubernetes with autoscaling, and sets up Evidently drift monitoring. The project runs in a partner org — not a notebook.
Which frameworks and tools will I use?
Core ML: Python, NumPy, Pandas, Polars, scikit-learn. Deep learning: PyTorch, TensorFlow, JAX, Hugging Face. MLOps: MLflow, Weights & Biases, DVC, Airflow, Prefect, Feast. Serving: Triton, BentoML, ONNX, SageMaker, Modal. Monitoring: Evidently, Arize.
Will I prep for AIPMM AI/Data Scientist and Pragmatic Principal ML Engineer certs?
Yes. The curriculum is mapped to the AIPMM AI/Data Scientist track and the Pragmatic Principal ML Engineer credential. We run two full mock exams and reimburse the voucher fee on first-attempt pass.
What's the time commitment per week?
Plan for 12–15 hours: 2 live classes × 2 hours, 1 lab × 3 hours training models on your dev tenant, and ~5 hours of project work (training pipelines, serving, drift). Saturday office hours with the TA team are optional, but most learners use them.
Is placement support really 1:1, and which companies hire data scientists / ML engineers?
Yes — a dedicated placement advisor from week 8, not a helpdesk. AI product hiring partners include Microsoft, Adobe, Salesforce, Atlassian, Notion, Linear, Anthropic, Hugging Face, Databricks, Snowflake, Stripe, Razorpay, Freshworks, Zoho, and Postman. Resume, LinkedIn, mock interviews, and warm intros are individual.
Online, weekend, or on-campus?
All three. On-campus at the Hyderabad flagship, live online (IST and PST cohorts), and a weekend track for working professionals. Every format ships the same shipped project — training pipeline, model serving, drift monitoring, LLM augmentation — only the schedule changes.
What if I fall behind, or can't continue mid-class?
Freeze your seat for up to 90 days and rejoin the next class — no extra fee. TAs run catch-up sessions every Saturday for anyone more than a week behind, and recordings of every live session are available for the lifetime of your account.
Still have a question? Talk to an advisor — no slides, no pitch.
Class ADS-023 starts 1 Jun 2026.
40 seats. 12 already claimed.
Book a 20-minute advisor call. We'll walk through the curriculum, match it to your current role, and show you two real projects from class 022.
