Industry Immersion Programme — Genvarsity

Industry Immersion Programme Mastery: Production Level 12 Months

Industry Immersion Programme

Twelve months of structured weekly curriculum at Advanced and Production level, running concurrently with a full-year company engagement. Five hours of live sessions per week. The remainder of each week is spent on live company work.

Talk to a Counsellor

Months Total

240+

Curriculum Hours

5 hrs

Weekly Live Sessions

Live

Remote Delivery

// programme structure

Two phases.
One continuous engagement.

The programme runs across two six-month phases, each with a defined curriculum depth and mastery level. Both phases run alongside a single, continuous host company engagement — from programme start through completion.

Phase 01 · Months 1–6

Applied AI at Advanced Level

Seven modules covering Machine Learning, Deep Learning, Data Science, Generative AI, Agents and Applied Experimentation. Modules are sequenced so that all three domains develop in parallel rather than as isolated blocks. Five hours of live sessions per week throughout.

Duration6 months

Curriculum~120 hours

Mastery LevelAdvanced

Weekly Sessions5 hours live

Phase 02 · Months 7–12

Production-Level Engineering

Seven modules covering MLOps, advanced GenAI engineering, AI system design, production data engineering, client delivery methodology and AI governance. Designed to be taught in direct context of the live company work the student is doing throughout Phase 2.

Duration6 months

Curriculum~120 hours

Mastery LevelProduction

Weekly Sessions5 hours live

// Phase 01 · Months 1–6 · Advanced

Seven modules.
Advanced mastery throughout.

Modules are sequenced so that ML, Data Science and Generative AI threads develop in parallel rather than completing one before the next begins. Every module has hands-on work from the first session.

Advanced Supervised & Unsupervised Learning20 hours · Weeks 1–4

Core ML

▼

Classical ML at production depth — covering advanced ensemble methods, Bayesian hyperparameter optimisation, imbalanced learning, model interpretability and calibration. Students completing this module can build, diagnose and deliver ML models in professional engineering contexts, with explicit attention to failure modes and decision quality.

Ensemble Methods & Optimisation

XGBoost internals — gradient approximation, second-order optimisation, regularisation terms

LightGBM — leaf-wise growth, GOSS sampling, EFB feature bundling

CatBoost — ordered boosting, categorical feature encoding

Bayesian hyperparameter optimisation — Gaussian processes, acquisition functions, Optuna TPE

Advanced feature engineering — target encoding, cyclic features, polynomial interactions

Imbalanced learning — SMOTE variants, cost-sensitive learning, threshold optimisation

Multi-label and multi-output learning frameworks

Evaluation, Interpretability & Unsupervised

Model calibration — Platt scaling, isotonic regression, reliability diagrams

SHAP values — TreeSHAP, KernelSHAP, interaction values

LIME and counterfactual explanations for stakeholder communication

Anomaly detection — Isolation Forest, LOF, One-Class SVM, autoencoder-based

Clustering — HDBSCAN, GMMs, optimal cluster validation

Dimensionality reduction — UMAP, kernel PCA, autoencoders for compression

scikit-learn Pipeline design for reproducible, production-oriented ML

Toolkit

scikit-learnXGBoostLightGBMCatBoostSHAPOptunaimbalanced-learn

Statistical Inference & Causal Reasoning15 hours · Weeks 3–7

Data Science

▼

Inferential statistics, experimental design and causal analysis at professional depth — covering the frameworks used in commercial analytics, product research and risk contexts. Students completing this module can design, run and correctly interpret experiments, including causal studies, in production environments.

Inference & Experimentation

Hypothesis testing — parametric, non-parametric, sequential testing

Multiple testing corrections — Bonferroni, Benjamini-Hochberg FDR

Bayesian inference — prior specification, posterior updating, MCMC intuition

A/B testing — peeking problem, power calculations, early stopping rules

CUPED variance reduction for experiment sensitivity

Multi-armed bandits — Thompson sampling, UCB, contextual bandits

Bayesian A/B testing — posterior distributions as decision tools

Causal Inference

Potential outcomes framework — ATE, ATT, treatment effect estimation

Directed acyclic graphs — confounders, colliders, mediators, d-separation

Propensity score matching and inverse probability weighting

Instrumental variables — identification strategy, weak instrument tests

Difference-in-differences — parallel trends assumption, staggered adoption

Regression discontinuity — sharp and fuzzy designs

Uplift modelling — S-learner, T-learner, X-learner for heterogeneous treatment effects

Toolkit

SciPystatsmodelsPyMCCausalMLDoWhy

Deep Learning Architecture & Training Dynamics20 hours · Weeks 5–11

Deep Learning

▼

Neural network design, training dynamics and the transformer architecture at working depth — covering advanced optimisation, efficient architectures, contrastive learning and transfer learning. The transformer walkthrough in this module is the direct prerequisite for the Generative AI modules that follow.

Architecture & Training

Architecture design principles — width vs depth, skip connections, residual learning

Advanced optimisers — AdaGrad, RMSProp, Adam variants, Lion, SAM (Sharpness-Aware Minimisation)

Learning rate schedules — cosine annealing, warm restarts, linear warmup

Advanced regularisation — DropBlock, mixup, CutMix, label smoothing

Normalisation — batch, layer, group and instance normalisation — selection criteria

Contrastive learning — SimCLR, MoCo, BYOL, self-supervised pretraining

Experiment tracking and reproducibility — W&B sweeps, MLflow

Transformer & Transfer Learning

Scaled dot-product attention — mathematical treatment and computational complexity

Multi-head attention, positional encoding, feed-forward sublayers, residual connections

Efficient transformers — sparse attention, linear attention, Flash Attention

Vision transformers (ViT) — patch embeddings, class tokens, hybrid architectures

Multi-modal architectures — CLIP, vision-language models

Transfer learning strategies — layer freezing, discriminative learning rates

Fine-tuning pre-trained models on custom datasets using Hugging Face

Toolkit

PyTorchHugging FacetimmWeights & BiasesMLflow

Data Engineering & Advanced Analytics15 hours · Weeks 8–12

Data Engineering

▼

Production data engineering and applied analytics — SQL at execution-plan level, cloud data warehouse architecture, dbt, pipeline orchestration and advanced time series. Equips students to work within and contribute to production data infrastructure in a professional context.

SQL & Data Warehousing

SQL at depth — query execution plans, index design, lateral joins, recursive CTEs

Window function patterns — frame specifications, RANGE vs ROWS

Dimensional modelling — star schema, SCD types, data vault concepts

dbt — models, tests, macros, snapshots, data contracts, packages

Cloud data warehouses — BigQuery partitioning and clustering, Snowflake virtual warehouses

Feature stores — offline/online stores, point-in-time correctness

Pipeline orchestration — Airflow DAG design, task dependencies, SLA management

Advanced Analytics & Time Series

Advanced time series — SARIMA, SARIMAX, Prophet with external regressors

Neural forecasting — LSTM, Temporal Fusion Transformer, N-BEATS

Hierarchical and multi-series forecasting

Anomaly detection in time series — STL decomposition, MSTL

Power BI — composite models, DAX calculation groups, deployment pipelines

Data observability — quality checks, schema evolution monitoring

Analytical communication — structuring findings for executive audiences

Toolkit

SQLdbtBigQueryAirflowProphetDartsPower BI

Generative AI & LLM Engineering20 hours · Weeks 10–17

Generative AI

▼

LLM architecture, prompt engineering and GenAI application development at engineering depth — covering the full pipeline from commercial API use through to advanced RAG architecture and systematic evaluation. Students completing this module can design, build and evaluate production-grade GenAI applications.

LLM Architecture & Prompt Engineering

LLM pretraining objectives — masked LM, causal LM, span corruption

RLHF pipeline — supervised fine-tuning, reward modelling, PPO

Tokenisation — BPE, WordPiece, SentencePiece, tokenisation artefacts

Embedding spaces — bi-encoders vs cross-encoders, embedding fine-tuning

Prompt engineering at engineering level — systematic design, DSPy for declarative prompting

Structured outputs — JSON schema enforcement, function calling, tool use patterns

Prompt failure analysis — hallucination taxonomy, refusal patterns, output drift

Advanced RAG & Retrieval

Document loading — parsing strategies, structure extraction, multi-modal documents

Chunking strategies — recursive, semantic, late chunking, parent-child hierarchy

Embedding model selection — evaluation methodology, domain adaptation

Hybrid retrieval — BM25 + dense, reciprocal rank fusion

Reranking — cross-encoders, ColBERT, multi-vector retrieval

Query expansion — HyDE, multi-query generation, step-back prompting

RAG evaluation — RAGAS, G-Eval, component-level assessment

Toolkit

OpenAI APIAnthropic ClaudeLangChainLlamaIndexDSPyPineconeWeaviateRAGAS

AI Agents & Agentic Systems15 hours · Weeks 15–20

Agents

▼

Design and implementation of agentic AI systems — covering architectures, tool use, multi-agent coordination and operational reliability. Students build functional agent systems using current frameworks with explicit attention to failure modes and production constraints.

Architecture & Tool Use

Agent architectures — ReAct, Plan-and-Execute, Reflexion, LATS

Tool use and function calling — schema design, error handling, retry strategies

Memory systems — short-term context management, episodic and semantic memory

LangGraph for stateful workflows — nodes, edges, conditional routing, checkpointing

MCP (Model Context Protocol) — server architecture, tool registration, session management

Browser and computer-use agents — web automation, structured data extraction

Multi-Agent Systems & Reliability

Multi-agent architectures — supervisor patterns, swarm coordination, handoffs

Agent communication — message passing, shared state, event-driven coordination

Reliability engineering — guardrails, circuit breakers, fallback chains

Human-in-the-loop integration — interrupt points, approval workflows

Cost management — token budgets, model routing by task complexity

Agent observability — span-level tracing, audit trails, cost attribution

Toolkit

LangGraphCrewAIAutoGenMCPLangSmith

Applied Projects & Research-to-Production Workflow15 hours · Weeks 19–24

Applied

▼

Integrates the Phase 1 curriculum through applied project work — experiment tracking, reproducibility, model documentation and the professional workflow for taking a result toward a deployable system. Aligned with the student's live company project where applicable.

Workflow & Reproducibility

Experiment tracking — MLflow Registry, W&B sweeps, systematic model comparison

Data versioning — DVC pipelines, dataset versioning, reproducible training runs

Research-to-production workflow — feasibility, PoC, pilot, production stage gates

Baseline establishment and model selection methodology

Model cards and documentation standards — Hugging Face, Google Model Card Toolkit

Git workflow for ML — branching strategy, code review, CI for notebooks

Communication & Portfolio

Technical writing for ML — architecture documents, experiment reports

Presenting ML results to non-technical stakeholders

Uncertainty communication — model limitations, confidence intervals

Portfolio development — GitHub structure, project documentation

Streamlit and Gradio for interactive project demonstrations

End-to-end applied project integrating Phase 1 modules

Toolkit

MLflowDVCW&BGitStreamlitGradio

Phase 02 — Production Level

Months 7–12 · ~120 hours · 5 hrs/week · Company engagement continues

// Phase 02 · Months 7–12 · Production

Seven modules.
Production depth throughout.

Phase 2 curriculum is structured to be directly applicable to the live company work the student is doing throughout. MLOps, advanced GenAI engineering, system design and delivery methodology are most effectively taught in the context of a real operating environment.

MLOps & Production ML Infrastructure20 hours · Weeks 25–31

MLOps

▼

The engineering discipline of deploying ML models to production and maintaining them operationally — covering serving architectures, containerisation, monitoring, pipeline orchestration and cloud ML platforms. Taught in direct context of the company engagement.

Serving & Infrastructure

ML system design — problem framing, data requirements, baseline establishment criteria

Model serving — REST with FastAPI, gRPC for high-throughput, batch inference patterns

Containerisation — Docker multi-stage builds, model packaging, dependency management

Kubernetes for ML — deployments, resource allocation, horizontal pod autoscaling

Feature pipelines — real-time computation, versioning, backfilling strategies

Cloud ML platforms — AWS SageMaker Pipelines, GCP Vertex AI, Azure ML

Pipeline orchestration — Airflow, Kubeflow Pipelines, Prefect

Monitoring, CI/CD & Production Management

Model monitoring — data drift (PSI, KS test), concept drift detection

Performance degradation monitoring — metric tracking, alerting thresholds

CI/CD for ML — GitHub Actions for training, evaluation gates, registry promotion

A/B testing in production — shadow deployment, canary releases, champion-challenger

Model retraining triggers — scheduled, drift-triggered, performance-triggered

Experiment registry — versioning, lineage, stage transitions

Incident response for ML systems — diagnosis, rollback, stakeholder communication

Toolkit

MLflowDockerKubernetesFastAPISageMakerEvidentlyGitHub Actions

Advanced GenAI Engineering & Production Systems20 hours · Weeks 27–35

GenAI Production

▼

Production engineering for Generative AI systems — fine-tuning methodology, self-hosted deployment, LLM observability, cost engineering and production guardrails. Students completing this module can fine-tune, deploy and operate LLM-powered systems at production scale.

Fine-Tuning & Deployment

Fine-tuning decision framework — when to fine-tune vs RAG vs prompting

LoRA — rank decomposition, alpha scaling, target module selection

QLoRA — 4-bit quantisation with LoRA, memory-efficient training

Dataset preparation — instruction formatting, data quality, deduplication

DPO — direct preference optimisation, dataset construction, training dynamics

Quantisation formats — GPTQ, AWQ, GGUF — trade-offs and selection criteria

Self-hosted deployment — vLLM (PagedAttention), TGI, Ollama

Observability, Cost & Guardrails

LLM observability — LangSmith, LangFuse, Arize Phoenix — span-level tracing and cost attribution

Caching strategies — exact match, semantic cache, prompt caching

Model routing — routing by task complexity, latency requirements, cost budgets

Token optimisation — prompt compression, context summarisation strategies

Production guardrails — PII detection, content moderation, output schema validation

Continuous evaluation — eval dataset maintenance, regression testing for LLM outputs

Incident response — detection, diagnosis, model rollback, stakeholder communication

Toolkit

TRLUnslothvLLMTGILangSmithLangFuseBraintrustRagas

AI System Design & Architecture20 hours · Weeks 31–38

System Design

▼

Architectural principles and system-level thinking for AI systems at production scale — data architecture, distributed training, model compression, real-time serving, security and reliability. Students completing this module can design AI systems that meet latency, throughput, cost and reliability requirements in enterprise environments.

Architecture & Scalability

End-to-end AI system architecture — components, data flows, failure modes mapping

Data architecture for AI — lambda vs kappa, event streaming, CDC

Real-time ML systems — low-latency prediction serving, online feature computation

Distributed training — data parallelism, model parallelism, ZeRO optimisation

AI-native product architecture — LLM integration patterns, streaming UI, structured output parsing

Multi-modal pipeline design — document intelligence, vision-language architectures

Cost architecture — GPU/CPU/TPU selection, spot instances, inference cost modelling

Compression, Security & Reliability

Model compression — knowledge distillation, structured and unstructured pruning

Quantisation-aware training vs post-training quantisation

Security for AI systems — prompt injection, adversarial inputs, data poisoning

Jailbreak mitigation — input filtering, output validation, sandboxing strategies

Reliability engineering — SLAs for AI endpoints, error budgets, graceful degradation

Load testing — throughput profiling, latency percentile analysis

Infrastructure as code — Terraform for reproducible AI environments

Toolkit

RayKafkaRedisFastAPIgRPCTerraformk6

Production Data Engineering & DataOps15 hours · Weeks 34–39

DataOps

▼

Data platform architecture and production data engineering — streaming pipelines, data quality frameworks, versioning and the analytical engineering layer. Taught in context of production data infrastructure encountered in the company engagement.

Streaming & Platform Architecture

Data lakehouse architecture — medallion (bronze/silver/gold), data mesh concepts

Apache Kafka — topics, partitions, consumer groups, exactly-once semantics

Spark Streaming and Flink — stateful stream processing, windowing strategies

Data versioning — Delta Lake, Apache Iceberg — time travel, schema evolution

ML data pipelines — ETL for feature engineering, training data curation at scale

Semantic and metrics layer — MetricFlow, dbt Metrics, LookML

Quality, Observability & Governance

Data quality engineering — Great Expectations, Soda, data contracts

Schema evolution — backward/forward compatibility, versioning strategies

Data observability — freshness, volume and schema drift detection

Column-level lineage — impact analysis, OpenLineage

dbt at production level — advanced macros, exposures, data contracts, packages

Analytical engineering — documentation standards, testing strategy, CI for data

Toolkit

KafkaSparkdbtDelta LakeGreat ExpectationsDVC

Client Engagement & Applied AI Delivery15 hours · Weeks 37–43

Delivery

▼

The professional discipline of scoping, delivering and communicating AI work in a client or organisational context — problem framing, business case construction, stakeholder management, project lifecycle and post-deployment management. Taught directly against the student's company engagement.

Scoping & Delivery

Problem framing — business problem decomposition, feasibility assessment methodology

Data availability assessment — audit methodology, quality scoring, gap analysis

AI project lifecycle — discovery, proof of concept, pilot, production stage gates

Success metrics design — business KPIs, technical metrics, guardrail metrics

Business case construction — ROI framework, cost-benefit analysis, payback period

Change management for AI — adoption planning, training design, resistance handling

Communication & Post-Deployment

Technical to non-technical translation — framing risk, uncertainty and trade-offs for leadership

Executive presentations for AI — pyramid structure, uncertainty communication

Delivery documentation — model cards, architecture documents, runbooks, API documentation

Knowledge transfer and handoff — team briefing, documentation walkthrough

Post-deployment management — monitoring SLAs, retraining triggers, escalation protocols

Responsible disclosure — bias findings, performance limitations, regulatory obligations

AI Governance, Regulation & Ethics in Practice15 hours · Weeks 40–46

Governance

▼

The regulatory and ethical framework within which production AI systems operate — EU AI Act, India's DPDP Act, algorithmic fairness methodology, privacy-preserving techniques and governance implementation. Taught at applied level using primary regulatory documents and case studies drawn from current deployments.

Regulation & Fairness

EU AI Act — risk classification, prohibited AI, conformity assessment, documentation obligations

India DPDP Act — data principal rights, consent mechanisms, cross-border transfer rules

GDPR implications for AI — automated decision-making, right to explanation

Algorithmic fairness — demographic parity, equalised odds, individual fairness, trade-offs

Bias audit methodology — source identification, measurement, documentation

Fairness toolkits — Fairlearn, Aequitas, IBM AI Fairness 360

Explainability in regulatory contexts — SHAP for compliance, counterfactual explanations

Privacy-Preserving ML & Governance

Federated learning — horizontal, vertical and federated transfer learning

Differential privacy — epsilon-delta privacy, DP-SGD, privacy budget management

Secure multi-party computation — concepts and enterprise application

AI governance frameworks — NIST AI RMF, ISO/IEC 42001, internal governance design

High-risk AI documentation — technical file, conformity declaration requirements

Incident response — detection, escalation, public disclosure, remediation

Human oversight design — meaningful human control, audit trail requirements

Toolkit

FairlearnAequitasSHAPIBM AIF360OpenDP

Production Capstone & System Integration15 hours · Weeks 44–48

Capstone

▼

Programme completion — integrating Phase 2 curriculum through a production-grade AI system aligned with the company engagement. Covers system testing, production readiness review, documentation and formal knowledge transfer. Reviewed by an industry panel before programme certificate is issued.

System Completion & Testing

End-to-end AI system implementation — ML pipeline and GenAI components integrated

Unit testing for ML — data validation, model behaviour contracts, output schemas

Integration testing — API contracts, end-to-end pipeline tests

Load and stress testing — throughput profiling, latency percentile analysis

Production readiness checklist — security review, monitoring coverage, documentation

Stakeholder sign-off — acceptance criteria, formal sign-off documentation

Documentation & Handoff

Architecture documentation — component diagrams, data flow diagrams, decision records

API documentation — OpenAPI specification, usage guides, versioning

Runbook development — operational procedures, escalation paths, known issues

Model cards — intended use, evaluation results, limitations, ethical considerations

Knowledge transfer — team briefing, documentation walkthrough

Industry panel review — technical and delivery assessment for programme completion

Toolkit

Pytestk6GitHub ActionsOpenAPIConfluence

// company engagement

Twelve months.
One continuous company engagement.

The company engagement begins at programme start and runs through completion. Five hours of weekly curriculum sessions run alongside. The remaining working time each week is spent at the host company on live project work.

Full-year host company placement

Company association from month one through month twelve.

Placement matching begins prior to programme start. Host company, supervisor and project scope are confirmed before the first session. The curriculum is designed to reinforce the company work throughout both phases.

Pre-startPlacement matching and host company confirmation. Project scope and supervisor assigned.

Month 1Company engagement commences. Phase 1 curriculum begins. Five hours of live sessions per week.

Month 7Phase 2 curriculum commences. Company engagement continues uninterrupted.

Month 12Production capstone reviewed by industry panel. Genvarsity IIP certificate and host company experience letter issued.

Stipend, if any, for the company engagement is decided at the sole discretion of the host company and does not form part of this programme's commitment. Stipend arrangements, where applicable, are communicated at the time of placement confirmation.

Host company experience letter — 12 months

Issued on the host company's letterhead, recording twelve months of engagement, project scope and supervisor details. A company document, independent of the Genvarsity programme certificate. Accepted by most universities under NEP-aligned internship credit regulations as the basis for awarding degree credits.

Curriculum aligned to company context

Phase 2 in particular is structured to be taught in direct context of the production work the student is doing at the host company. The five weekly session hours are applied to current engineering problems, not simulated scenarios.

Genvarsity IIP Certificate

Issued on programme completion, recording successful completion of both phases — 240+ hours of curriculum at Advanced and Production level — along with the industry panel review outcome for the production capstone.

// admissions · 2026

Programme details are
discussed directly.

Speak with a Genvarsity counsellor about the programme structure, the company placement process and the admission process for the 2026 cohort.

Talk to a Counsellor

Industry Immersion Programme

Two phases.One continuous engagement.

Seven modules.Advanced mastery throughout.

Seven modules.Production depth throughout.

Twelve months.One continuous company engagement.

Programme details arediscussed directly.

Two phases.
One continuous engagement.

Seven modules.
Advanced mastery throughout.

Seven modules.
Production depth throughout.

Twelve months.
One continuous company engagement.

Programme details are
discussed directly.