Glossary

Common AI terms in plain English, with the Hindi term in brackets where one exists.

Accuracy (सटीकता): How often a model's answers are correct out of all the answers it gave. High accuracy can still hide mistakes on rare cases.
Agent (एजेंट): An AI system that can take steps on its own, such as searching, calling tools, or writing files, to finish a task. Agents need clear limits because a wrong step can have real effects.
AI (कृत्रिम बुद्धिमत्ता): Software that performs tasks that used to need human thinking, such as writing, answering questions, or recognising images. Most AI you meet today learns patterns from large amounts of data.
API: A way for two pieces of software to talk to each other. Apps use APIs to send a request to an AI model and get its answer back.
Bias (पूर्वाग्रह): A pattern in a model's output that unfairly favours or disadvantages a group. It usually comes from the data the model learned from.
Chatbot (चैटबॉट): A program you talk to in a chat window, often powered by a language model. It replies in text and can also handle files or images.
Context window (संदर्भ विंडो): The amount of text a model can consider at one time, including your question and the conversation so far. When it fills up, older parts are dropped or summarised.
Credits (क्रेडिट): Free units used to start assessments and get AI advisor replies. You get 100 when you join and 10 more every day. Neura Pro assessments use no credits.
DPDP Act (डिजिटल व्यक्तिगत डेटा संरक्षण अधिनियम): India's Digital Personal Data Protection Act, 2023. It sets the rules for collecting and using personal data, and gives people rights over their data.
Embedding (एम्बेडिंग): A list of numbers that stands for the meaning of a piece of text, image, or audio. Similar meanings sit closer together, which is what powers search and recommendations.
Fine-tuning (फाइन-ट्यूनिंग): Extra training that adjusts a ready-made model for a specific task or style. It is usually smaller and cheaper than training a model from scratch.
Guardrails (गार्डरेल): Rules and checks placed around an AI system to keep its answers safe, on topic, and within policy. They catch problems before a user sees them.
Hallucination (मनगढ़ंत जवाब): When a model states something false as if it were true. It happens because the model predicts likely text rather than looking up facts.
LLM (बड़ा भाषा मॉडल): A large language model, trained on huge amounts of text to write and understand language. ChatGPT and similar tools are built on LLMs.
Model (मॉडल): The learned part of an AI system: patterns picked up from training data, used to make predictions or produce answers. A model is not a database of facts.
Neura: NeuraCamp's free plan: your score, your level, a full answer review, your top strengths and your first step.
Neura Pro: NeuraCamp's paid plan, ₹99 once for 12 months with no auto-renewal. It adds the full readiness report, a week-by-week plan, projects, role matches and a NeuraCamp certificate.
Overfitting (ओवरफिटिंग): When a model learns its training data too closely, including its noise, and then performs poorly on new data. It looks accurate in testing but does not generalise.
Percentile (प्रतिशतक): Your rank among people who took the same assessment. The 80th percentile means you scored higher than 80 out of 100 of them. NeuraCamp shows it only once enough people have taken an assessment.
Personal data (व्यक्तिगत डेटा): Any information that can identify a person, such as a name, phone number, photo, or account details. It needs care when you paste it into AI tools.
Precision (परिशुद्धता): Of the items a model marked as positive, the share that really were positive. It answers: when the model says yes, how often is it right?
Prompt (प्रॉम्प्ट): The instruction or question you give an AI tool. Clear goals and useful context usually produce better answers.
Prompt injection (प्रॉम्प्ट इंजेक्शन): Hidden instructions placed in text or a file that try to make an AI system ignore its real rules. Treat outside content as untrusted and keep guardrails in place.
RAG (रिट्रीवल ऑगमेंटेड जनरेशन): Retrieval-augmented generation: the system first fetches relevant documents, then asks the model to answer using them. It grounds answers in real sources and reduces made-up ones.
Recall (रिकॉल): Of all the items that should have been found, the share the model actually found. It answers: how many real cases did it miss?
Skill area (कौशल क्षेत्र): One of the six groups NeuraCamp uses to score your answers, such as prompting or security and governance. Your results show a score for each area.
Token (टोकन): A small chunk of text, roughly a word part, that a model reads and writes. Usage limits and costs are usually counted in tokens.
Training data (ट्रेनिंग डेटा): The examples a model learns from. The quality and fairness of this data shape what the model can do and where it fails.