AI Skills & Technologies Map
Nine disciplines that define working AI in 2026 — the skills that compound and the technology picks that ship. Examined, corrected and rebuilt as an ImageFirm reference sheet.
Plan view
One core, nine load-bearing disciplines. Each tile opens its sheet.
Machine Learning
The statistical foundation every other branch stands on.
Skills
- Data preprocessing & feature engineering
- Model selection & training
- Hyperparameter tuning
- Evaluation metrics (AUC, F1, calibration)
- Model deployment & MLOps
- Drift detection & monitoring
Technologies
- scikit-learn
- XGBoost / LightGBM
- PyTorch
- MLflow
- Weights & Biases
- Ray
Deep Learning
Representation learning at scale — architectures, training, hardware.
Skills
- Neural network architecture design
- Convolutional & recurrent networks
- Attention & transformer models
- Transfer learning & fine-tuning (LoRA / QLoRA)
- Distributed GPU / TPU training
- Multimodal model training
Technologies
- PyTorch Lightning
- JAX
- Hugging Face Transformers
- DeepSpeed / FSDP
- CUDA / cuDNN
- ONNX / TensorRT
Natural Language Processing
From raw text to grounded, cited answers.
Skills
- Text preprocessing & embeddings
- Named entity recognition & extraction
- Machine translation & summarization
- Semantic search & RAG pipeline design
- Knowledge graph construction
- Conversational AI design
Technologies
- sentence-transformers
- spaCy
- LlamaIndex / LangChain
- Elasticsearch
- Pinecone / Weaviate / pgvector
- Rerankers (Cohere, Voyage)
Generative AI
Creation across text, image, audio and video — with provenance.
Skills
- Prompt design & chaining
- Fine-tuning foundation models
- Multimodal integration (text + image + audio)
- Synthetic data generation
- Style transfer & personalization
- Responsible generation & provenance (C2PA)
Technologies
- Claude / GPT / Gemini
- Llama / Mistral
- Midjourney / Flux
- Stable Diffusion
- Runway / Veo
- ElevenLabs / Suno / Udio
Agentic AI
Systems that plan, act and verify — under human authority.
Skills
- Agent design patterns (planner–executor, hierarchical, reflection)
- Reasoning & planning algorithms
- Goal decomposition & task orchestration
- Memory management (episodic & semantic)
- Tool use & multi-agent protocols
- Agent evals, benchmarking & human-in-the-loop gates
Technologies
- Claude Agent SDK
- OpenAI Agents SDK
- LangGraph
- CrewAI / AutoGen
- Model Context Protocol (MCP)
- Computer-use & browser agents
Computer Vision
Machine perception: detect, segment, reconstruct, track.
Skills
- Image preprocessing & augmentation
- Object detection & recognition
- Semantic & instance segmentation
- 3D vision & reconstruction (NeRF, Gaussian splatting)
- Video analytics & tracking
- AR / VR integration
Technologies
- OpenCV
- YOLO (v8–11)
- SAM 2
- CLIP
- MediaPipe
- NVIDIA DeepStream
Robotics & Autonomous Systems
Intelligence with consequences in the physical world.
Skills
- Motion planning & SLAM
- Sensor fusion (lidar, radar, vision)
- Reinforcement & imitation learning
- Real-time control systems
- Human–robot interaction
- Multi-robot coordination
Technologies
- ROS 2
- Isaac Sim / Isaac Lab
- Gazebo
- MuJoCo / Gymnasium
- Boston Dynamics SDK
- VLA foundation models (GR00T, π0)
AI Ethics & Governance
The license to operate — engineered, not declared.
Skills
- Bias detection & mitigation
- Explainable AI (XAI) design
- Privacy-preserving AI (federated learning, differential privacy)
- Auditing & compliance (EU AI Act, NIST AI RMF)
- Red-teaming & risk analysis
- Policy frameworks for responsible AI
Technologies
- AI Fairness 360
- SHAP / LIME
- Fairlearn
- Model & system cards
- Eval harnesses (HELM, LM Eval, promptfoo)
- Guardrail engines (NeMo Guardrails, Llama Guard)
Local & Private AI
Your data, your weights, your infrastructure.
Skills
- Model quantization (GGUF, 4-bit)
- On-device & edge deployment
- Self-hosted inference & serving
- Data-sovereignty architecture
- Hybrid local / cloud routing
- Cost & latency engineering
Technologies
- Ollama
- llama.cpp
- vLLM
- LM Studio
- MLX (Apple silicon)
- Open-weight models (Llama, Mistral, Qwen, Gemma)
Examiner's note
Three corrections to the version circulating on your feed. One — it lists “AI Agents” and “Agentic AI” as separate branches carrying the same six skills; that is one discipline, merged here as 05. Two — the 2023 relics are retired: AutoGPT, BabyAGI and HuggingGPT gave way to what actually ships in 2026 — agent SDKs, LangGraph, CrewAI and MCP. Three — the branch every version of this map forgets: Local & Private AI. If sovereignty over data and weights is not on your map, someone else is drawing it for you.
Uzay Kadak — ImageFirm · Rev 2026.08