Track A · Deep AI for Computer Science

Track A

An 8-course AI specialization for B.Tech CSE/IT/AI-ML/DS, BCA, and M.Tech students at Indian private universities. From classical ML to agentic AI systems and AI engineering. NEP-aligned, 24–30 credits.

Talk to our partnerships team Download the Track A curriculum

Track A curriculum snapshot — eight courses A1 through A8.

NEP 2020Credit-mapped NAACEvidence-ready AICTEModel-curriculum aligned NBAOBE-mapped

The course list

8 courses in Track A.

Each module is co-evaluated by Kompas resident faculty and your university's academic council. Module content refreshes every six months.

Foundations of AI & ML

Sem 3–43 credits

Skill outcomes

Apply linear algebra, probability, and statistics to machine learning problems
Implement classical ML algorithms in scikit-learn end-to-end
Evaluate models using appropriate metrics and the bias-variance tradeoff
Run principled model selection and hyperparameter tuning workflows

Deep Learning & Neural Networks

Sem 4–54 credits

Skill outcomes

Build CNNs, RNNs, and transformer models in PyTorch 2.x
Train networks at scale with regularization and modern optimization
Apply transfer learning to adapt pretrained models to new tasks
Operate GPU training workflows in the on-campus Kompas AI Studio

Prerequisites: A1

Natural Language Processing & LLMs

Sem 5–64 credits

Skill outcomes

Derive transformer architectures from tokenization and attention first principles
Fine-tune small open-weight models such as Llama- and Mistral-class LLMs
Distinguish pretraining, instruction tuning, and RLHF in practice
Design evaluation suites for NLP and LLM tasks

Prerequisites: A2

Generative AI & Diffusion Models

Sem 63 credits

Skill outcomes

Train and condition diffusion models including Stable Diffusion and ControlNet
Reason about diffusion math intuition across image, audio, and video modalities
Evaluate generative output with task-appropriate quality and safety metrics
Implement watermarking and provenance for generated content

Prerequisites: A2

Agentic AI Systems & LLMOps

Sem 6–74 credits

Skill outcomes

Build deployable agentic applications using LangChain, LangGraph, and MCP
Architect RAG pipelines over vector databases like pgvector, Qdrant, and Weaviate
Instrument agent observability, evaluation, and cost management
Apply orchestration patterns for multi-step tool use in production

Prerequisites: A3

AI Engineering & Deployment

Sem 74 credits

Skill outcomes

Serve models in production with FastAPI, BentoML, and Kubernetes
Deploy workloads on AWS Bedrock/SageMaker, Azure AI Foundry, and GCP Vertex AI
Run prompt management, A/B testing, and gradual rollouts for AI features
Operate observability stacks tailored to ML and LLM services

Prerequisites: A5

Responsible AI, Safety & Alignment

Sem 72 credits

Skill outcomes

Assess models for bias and fairness across deployment contexts
Run red-teaming exercises against LLM and agentic systems
Map products to the EU AI Act, India's DPDP Act, and sectoral compliance
Publish model cards and system cards documenting alignment posture

Prerequisites: A5

Industry-Sponsored AI Product Build (Capstone)

Sem 86 credits

Skill outcomes

Scope and design an AI product with a sponsoring industry partner
Build and deploy a working AI service used by the sponsoring company
Document the system with a model card, system card, and runbook
Defend the product in co-evaluation by university faculty and industry sponsor

Prerequisites: A6 → A7

Skill Score on exit

Mastery is measured, not assumed.

Every Track A graduate exits with a verifiable Kompas AI Skill Score, calibrated against an industry-co-signed rubric and applied identically across every campus we operate on.

Foundational
Practitioner
Applied
Advanced
Expert

Track A graduates typically exit between Practitioner and Applied. Industry-sponsored capstone work moves the strongest students into the Advanced band.

Capstones from this track

What a Track A student ships at the end.

Examples below are illustrative until real partner-cohort capstones land. The shape — sponsor, problem, outcome, public artefact — does not change.

See the full capstone library

Track AHealthcare · Indic NLPFoundation-model lab

Vernacular speech-to-text for rural ASHA workers

Problem

ASHA workers record patient interactions in code-switched Hindi and English. Off-the-shelf Whisper-class models hallucinate medical entities and miss Indic phonology, making the transcripts unreliable for clinical follow-up.

Target outcome

A domain-adapted Whisper-derivative reaching >90% medical-term accuracy on a held-out 1,200-utterance evaluation set, with a privacy-by-design audit. Deployed at three primary health centres for a four-week field pilot.

Track ALegal · RAGEnterprise AI consultancy

Production RAG for legal research at a state High Court

Problem

Court researchers need to retrieve precedent across 1.2M judgments without hallucinated citations — the failure mode that disqualified prior LLM trials.

Target outcome

A hybrid retrieval pipeline (dense + BM25) with a citation-verification step that refuses to answer when retrieval confidence is below threshold. Zero hallucinated citations across a 500-query evaluation set, defended in a registrar-style audit.

See 4 more Track A capstones →

Other tracks