artificial intelligence · industry roadmap 2025

artificial intelligence · 2025

deep learning

agents · llm

computer vision

Phase 1 · foundations of AI

history, intelligent agents, search, CSPs, knowledge representation, planning – weeks 1–8

History & philosophy

Turing test, Dartmouth, AI winters
expert systems, statistical revolution
transformer era, AGI debate

Intelligent agents

PEAS, environment types
reflex, goal‑based, utility agents
learning agents

Search algorithms

BFS, DFS, A*, heuristics
minimax, alpha‑beta pruning
MCTS, AlphaZero

CSP & planning

backtracking, AC‑3, heuristics
STRIPS, PDDL, HTN
MDPs, situation calculus

Phase 2 · learning in AI systems

Learning paradigms

supervised, unsupervised, RL
self‑supervised, few‑shot, transfer
federated, continual learning

Neural network theory

universal approx, representation
loss landscapes, scaling laws

Reinforcement learning

MDP, Bellman, Q‑learning
DQN, PPO, RLHF

PyTorch & training

autograd, nn.Module, DataLoader
mixed precision, distributed

Phase 3 · deep learning architectures

CNNs & vision

convolution, pooling, ResNet
EfficientNet, ViT, SAM

Transformers

self‑attention, multi‑head, QKV
BERT, GPT, T5, scaling

Generative models

VAE, GANs, diffusion (DDPM, SD)
Stable Diffusion, DALL‑E

Graph neural nets

GCN, GAT, message passing
AlphaFold, drug discovery

Phase 4 · NLP, LLMs & generative AI

LLM foundations

tokenization, pre‑training, emergent
RLHF, DPO, constitutional AI

Prompt engineering

zero/few‑shot, CoT, ReAct
system prompts, structured output

RAG & vector dbs

chunking, embeddings, hybrid search
Pinecone, Chroma, pgvector

AI agents

tool calling, ReAct, LangGraph
multi‑agent (CrewAI, AutoGen)

Phase 5 · computer vision & perception

Object detection

YOLO, Faster R‑CNN, DETR
NMS, IoU, mAP

Segmentation

semantic, instance, SAM
Mask R‑CNN, SAM2 video

3D vision & depth

monocular depth, NeRF, Gaussian
point cloud (PointNet)

Autonomous systems

sensor fusion, BEV, prediction
RT‑2, Pi0, sim‑to‑real

Phase 6 · AI systems engineering & MLOps

Training infra

GPU clusters, distributed (FSDP)
mixed precision, gradient ckpt

Model compression

quantisation (GPTQ, AWQ)
pruning, distillation, ONNX

LLM serving & MLOps

vLLM, continuous batching, Triton
MLflow, Evidently, CI/CD

Phase 7 · AI safety, ethics & alignment

Alignment & safety

outer/inner alignment, reward hacking
constitutional AI, red teaming

Fairness & bias

bias types, fairness metrics
disparate impact, audit tools

Governance & XAI

EU AI Act, model cards
SHAP, mechanistic interpretability

structured weekly roadmap

Phase	Timeline	Core topics	Deliverable
Phase 1	Weeks 1‑8	history, agents, search, CSP, knowledge, planning	implement A*, minimax, Bayes net
Phase 2	Weeks 9‑20	ML paradigms, neural theory, RL, PyTorch	classifier + RL agent
Phase 3	Weeks 21‑34	CNNs, transformers, GANs, diffusion, GNNs	image classifier + sentiment model
Phase 4A	Weeks 35‑46	LLMs, prompting, RAG, agents, fine‑tuning	RAG chatbot + agent
Phase 4B	Weeks 35‑46	CV: YOLO, SAM, depth, autonomous systems	custom detector + pipeline
Phase 5	Weeks 47‑54	distributed training, compression, serving, MLOps	deployed AI system + monitoring
Phase 6	Weeks 55‑60	alignment, fairness, governance, interpretability	fairness audit + safety eval
Phase 7	Weeks 55‑65	specialisation: agents / robotics / safety	deep capstone project

⚡ industry‑level capstone tracks

AI agent system

LangGraph multi‑agent: research + writer + fact‑checker
tools: web search, code exec, RAG memory
safety gates, streaming UI

multimodal medical AI

vision‑language model (LLaVA) + report generation
QLoRA fine‑tuning, GradCAM explainability
physician review gate, confidence thresholds

pytorch 2.x huggingface langgraph vLLM kubeflow evidently

explore capstone structure

must‑have (first job)

python, pytorch, transformers
implement attention from scratch
3+ deployed AI projects
LLM APIs, RAG, basic agents

top free resources

russell & norvig (AIMA)
fast.ai / karpathy's lectures
huggingface course / cs324
anthropic alignment research

interview focus

attention / backprop derivation
bias‑variance, scaling laws
AI system design + safety