Types of AI Tasks — Classification, Generation, Retrieval

# AI Task Types — Which Architecture Fits Your Problem?

# ─────────────────────────────────────────────────────────────────
# 1. DISCRIMINATIVE TASKS — Input → Label/Number
# ─────────────────────────────────────────────────────────────────
discriminative_tasks = {
    "Image Classification":    ("CNN, ViT",           "Is this a cat or dog?"),
    "Object Detection":        ("YOLO, DETR",          "Where are the pedestrians?"),
    "Sentiment Analysis":      ("BERT, DistilBERT",     "Is this review positive?"),
    "Named Entity Recognition":("BERT + token classifier","Find company names in text"),
    "Tabular Prediction":      ("XGBoost, TabNet",      "Will this customer churn?"),
    "Speech Recognition":      ("Whisper",              "Convert audio → text"),
}

# ─────────────────────────────────────────────────────────────────
# 2. GENERATIVE TASKS — Input → New Content
# ─────────────────────────────────────────────────────────────────
generative_tasks = {
    "Text Generation":         ("GPT-4, Llama 3",       "Write an email given a prompt"),
    "Image Generation":        ("Stable Diffusion",     "Generate photo from text"),
    "Code Generation":         ("Codex, DeepSeek Coder","Write function from docstring"),
    "Translation":             ("T5, mBART",            "English → French"),
    "Summarization":           ("BART, Pegasus",        "Summarize a news article"),
    "Text-to-Speech":          ("ElevenLabs, XTTS",     "Convert text to natural speech"),
    "Music Generation":        ("MusicGen",             "Generate music from description"),
}

# ─────────────────────────────────────────────────────────────────
# 3. RETRIEVAL TASKS — Query → Ranked Relevant Items
# ─────────────────────────────────────────────────────────────────
retrieval_tasks = {
    "Semantic Search":         ("FAISS + Embeddings",   "Find similar documents"),
    "RAG":                     ("LangChain + ChromaDB", "Q&A over your own documents"),
    "Recommendation":          ("Matrix Factorization", "Suggest products/content"),
    "Re-ranking":              ("Cohere Rerank, BGE",   "Reorder search results by relevance"),
}

# ─────────────────────────────────────────────────────────────────
# DECISION GUIDE — Which task type is YOUR problem?
# ─────────────────────────────────────────────────────────────────
def identify_task_type(problem_description: str) -> str:
    if any(w in problem_description.lower() for w in ["classify", "detect", "predict", "score"]):
        return "DISCRIMINATIVE — use BERT, CNN, XGBoost"
    elif any(w in problem_description.lower() for w in ["generate", "write", "create", "synthesize"]):
        return "GENERATIVE — use GPT, Diffusion, T5"
    elif any(w in problem_description.lower() for w in ["find", "search", "retrieve", "recommend"]):
        return "RETRIEVAL — use Embeddings + Vector DB"
    return "HYBRID — combine approaches with RAG + LLM"

# Examples:
problems = [
    "classify customer emails into 5 categories",
    "generate marketing copy from product specs",
    "find the most relevant FAQ answer for a user question",
    "answer questions from a 500-page PDF document",
]

for p in problems:
    task_type = identify_task_type(p)
    print(f"Problem: '{p}'")
    print(f"  → {task_type}\n")