AI · LESSON 18 / 25 · English
AI Tools & Frameworks
Choose a tool for the job, not a permanent winner.
Content reviewed: 22 October 2026 · தமிழில் படிக்க ↗
Understand it in plain language
An AI stack can include data tools, modelling libraries, model providers, retrieval components, evaluation, and monitoring. scikit-learn supports many classical ML tasks; PyTorch supports tensor computation and neural-network training; the Hugging Face ecosystem supports model and dataset workflows. A hosted model API is different from an orchestration framework. Exact versions, model availability, prices, and licences change, so check official documentation before choosing a deployment.
How the AI approach works
Select tools by the job they perform. Data preparation handles collection, cleaning, labels, and versioning. Modeling libraries fit or run models. Model hubs distribute artifacts and documentation. Hosted inference providers expose remote models. Retrieval components index and search documents. Orchestration connects steps, while evaluation and monitoring measure behavior. One framework rarely replaces this whole stack.
For tabular classification, scikit-learn is a practical place to compare classical models and pipelines. For custom differentiable models, PyTorch supplies tensor computation and automatic differentiation. Hugging Face tools support pretrained model, tokenizer, and dataset workflows. A vector index stores searchable numerical representations; it does not by itself generate answers or check whether a document is true.
Compare hosted and local inference against concrete requirements: quality on the task, Tamil behavior, latency, throughput, memory, data handling, model and dataset licences, operational effort, and total cost. Open weights do not mean unrestricted rights, zero cost, or guaranteed privacy. Hosted APIs do not remove the need for evaluation and application controls.
Avoid framework-first design. Prototype the smallest complete path, keep inputs and outputs explicit, pin tested versions, and record model revisions. Use adapters where provider changes are likely. Tool names, prices, and capabilities change; verify current official documentation rather than treating a course shortlist as a permanent ranking.
Break the idea into steps
- List the task, language needs, data constraints, and available hardware.
- Compare candidate tools with the same small evaluation set.
- Record versions and inspect model cards, dependencies, and licence terms.
Worked example: connect the method to the task
For a bilingual course FAQ, start with approved documents, a simple retriever, one evaluated model, a response format with source links, and a small test set. Add a vector index only if retrieval tests justify it; add an orchestration framework when the control flow actually benefits.
For predicting numeric enrollment demand, the same chat stack may be unnecessary. A versioned table, a baseline forecast, and a classical model can be a more suitable experiment.
Apply the idea: check your reasoning
Think before revealing the answer
Does installing a vector database create a complete RAG assistant?
Reveal the answer and reasoning
No. You still need ingestion, suitable representations, retrieval evaluation, permissions, generation with evidence, answer evaluation, and an operating workflow.
A common mistake: Open weights do not automatically mean unrestricted use, open training data, or zero deployment cost.
Before you move on
- Explain the main idea without repeating the technical label.
- Identify the approach, its assumptions, and the evidence needed to trust the result.
- Complete the challenge, then compare your reasoning with the answer.
Sources & a deeper next step
The refresh uses primary papers and official documentation. These optional links are in English; you can understand this lesson without reading them first. Product versions and availability can change.