Ai usecases 2025:

1. Perception What it does: Extract features from images/videos. Industries: Healthcare (medical imaging), Automotive (autonomous driving), Security (surveillance). Tech: CNNs, Vision Transformers (ViT), Image Segmentation (U-Net); Deep Learning (Computer Vision).

2. Speech Recognition What it does: Convert spoken audio to text. Industries: Customer Support, Consumer Electronics (voice assistants), Automotive (voice commands). Tech: RNNs, LSTMs, Transformers, CTC loss; Deep Learning (Speech Processing/NLP).

3. Text Understanding What it does: Comprehend text intent, entities, sentiment. Industries: Finance (document analysis), Legal (contract review), Customer Service. Tech: Transformers (BERT, RoBERTa), Named Entity Recognition (NER); Deep Learning (NLP).

4. Text Generation What it does: Produce coherent language output. Industries: Marketing (content creation), Media (summarization), Education (tutoring). Tech: Autoregressive Transformers (GPT family), Seq2Seq models; Deep Learning (NLP).

5. Knowledge Retrieval What it does: Retrieve relevant external info for tasks. Industries: Tech Support, Research, Healthcare. Tech: Dense vector retrieval with k-NN, embedding models (BERT embeddings), combined with LLMs (RAG); ML + DL (Information Retrieval + NLP).

6. Multimodal Fusion What it does: Align and integrate multiple data types. Industries: Retail (visual search), Entertainment (video captioning), Autonomous Systems. Tech: Multimodal Transformers, Cross-Attention; Deep Learning (Multimodal AI).

7. Prediction What it does: Forecast or detect anomalies from data. Industries: Finance (fraud detection), Manufacturing (predictive maintenance), Energy (demand forecasting). Tech: Regression, Random Forest, Gradient Boosting, LSTM; Machine Learning + Deep Learning (Time-Series Analysis).

8. Decision Making What it does: Optimize actions/plans based on goals. Industries: Logistics (route planning), Robotics, Gaming. Tech: Reinforcement Learning (Q-learning, Policy Gradients), Heuristic Search; Machine Learning (Reinforcement Learning).

9. Generative Content Creation What it does: Create new images, audio, code, etc. Industries: Advertising, Software Dev, Arts & Music. Tech: GANs, Diffusion Models, Autoregressive models (Codex); Deep Learning (Generative Models).

10. Autonomous Agents What it does: Autonomous perception, reasoning, and action. Industries: Autonomous Vehicles, Virtual Assistants, Industrial Automation. Tech: Integration of CNNs, Transformers, RL, Planning Algorithms; AI Systems + ML + DL (Agent-based AI).

Different section in AI:

Artificial Intelligence (AI)

AI is the broad field of creating intelligent systems that can mimic human behavior.

1. Symbolic AI / Classical AI

• Rule-Based Systems
• Knowledge Graphs
• Expert Systems

2. Machine Learning (ML)

ML is a subset of AI focused on systems that learn from data.

2.1 Learning Paradigms

• Supervised Learning

  • Tasks: Regression, Classification
  • Algorithms:
    • Linear Regression
    • Logistic Regression
    • Decision Trees
    • Random Forest
    • Support Vector Machine (SVM)

• Unsupervised Learning

  • Tasks: Clustering, Dimensionality Reduction
  • Algorithms:
    • K-Means
    • DBSCAN
    • PCA
    • t-SNE

• Semi-Supervised Learning

• Self-Supervised Learning

• Reinforcement Learning (RL)

  • Algorithms:
    • Q-Learning
    • SARSA
    • Deep Q-Network (DQN)
    • Proximal Policy Optimization (PPO)
    • A3C, DDPG, etc.

2.2 ML Techniques

• Classical ML – Uses above algorithms
• Deep Learning (DL) – Uses neural networks:
  • Feedforward Neural Networks (FNN / DNN)
  • Convolutional Neural Networks (CNN)
  • Recurrent Neural Networks (RNN)
    • LSTM, GRU
  • Autoencoders (AE, VAE)
  • Generative Adversarial Networks (GANs)
  • Transformers
    • Used in NLP, LLMs, Vision, Speech

---

Application Domains

3.1 Natural Language Processing (NLP)

Processes and understands human language.

• Tasks:

  • Text Classification
  • Named Entity Recognition (NER)
  • Machine Translation
  • Summarization, QA

• Models:

  • RNN, LSTM
  • BERT, RoBERTa
  • GPT Series
  • T5, XLNet

• Includes: Large Language Models (LLMs)

  • GPT-3, GPT-4, GPT-4o
  • LLaMA, Claude, PaLM
  • Used in chatbots, agents, RAG systems

3.2 Computer Vision (CV)

Processes and understands visual data (images, videos).

• Tasks:

  • Image Classification
  • Object Detection
  • Image Segmentation
  • Image Generation

• Models:

  • CNNs: VGG, ResNet, EfficientNet
  • Vision Transformers (ViT, DINO)
  • GANs: StyleGAN, CycleGAN

• Applications:

  • Facial Recognition, OCR, Medical Imaging

3.3 Speech / Audio Processing

Processes audio and speech.

• Tasks:

  • ASR (Automatic Speech Recognition)
  • TTS (Text-to-Speech)
  • Speaker Identification

• Models:

  • RNNs, CNNs
  • Transformers (e.g., Whisper)
  • WaveNet, Tacotron

3.4 Multimodal AI

Combines multiple input types: text + image + audio + video.

• Examples:
  • CLIP (text + image)
  • Whisper (speech + text)
  • Flamingo, GPT-4o, Gemini, Sora

3.5 Retrieval-Augmented Generation (RAG)

Combines LLMs with external data sources.

• Components:

  • Embedding Models
  • Vector Databases (e.g., FAISS, Pinecone)
  • LLMs for answer generation

• Use Cases:

  • Chat over documents
  • Internal knowledge bots
  • QA over web, PDFs, databases

NLP vs LLM (Brief)

• NLP is the science of understanding and working with language.
• LLMs are advanced tools (like ChatGPT) used within NLP to understand and generate text.
• Gen AI is the bigger umbrella that includes LLMs and also tools that make:
  • Images (like DALL·E, Midjourney)
  • Music (like Suno)
  • Videos (like Sora)
  • Code (like GitHub Copilot)

RAG Deep dive:

AI agents deep dive:

MCP server:

High level understanding of multi modal:

LLM architecture deep dive: