AI Training Assistant

An AI training assistant can prepare practice questions, scenario drafts and answer keys from approved course material. Tie each proposed answer to a source passage and have a trainer check relevance, ambiguity and coverage before using the exercise.

This AI Flowi Workflow guide designs a trainer-reviewed practice pack. It does not certify competence, grade employees or promise a learning outcome.

Choose a practice pack, not an assessment verdict

A document question-answering flow answers a question already supplied. A training draft creates a new exercise, so the reviewer must check both the question and its answer key. The source may support a fact without supporting the invented scenario’s assumptions.

Begin with an approved learning objective and a bounded source section. Specify who the exercise is for and what participants should practise. Use the material’s actual procedure and exceptions rather than asking a model to generate a general quiz about the whole organisation.

Proposed source-backed exercise contract
Exercise field	Source requirement	Trainer review
learning_objective	Objective selected by the trainer.	Check the exercise practises that objective.
scenario_question	Conditions consistent with the approved procedure.	Remove unsupported situation details.
answer_key	Answer or action supported by the supplied passage.	Check exceptions and acceptable alternatives.
source_passage / revision	Literal relevant material and its approved revision.	Confirm the answer can be verified.
ambiguity_note	Unresolved source or scenario interpretation.	Revise or withhold the exercise.
trainer_review	Human disposition after checking the draft.	Keep approval separate from model generation.

Prepare course sources and provider access

You need a workflow account, approved course material, a selected objective and a trainer responsible for checking exercises. Use your own embedding-provider configuration for the searchable knowledge base. The generation step also needs your own provider account and API key, with provider usage charged separately.

Keep source revisions and applicable procedures visible in the documents. Separate material intended for different roles when their permitted actions differ. Do not include individual employee performance records in the practice-source base. The knowledge-base documentation covers ingestion and chunk review; this page adds exercise-generation and answer-key controls.

Configure a practice-material draft

example configuration, not a recorded test

The proposed training pack is documented in the catalogue; not production-tested by Ai Flowi.

Knowledge Base: use the course retrieval setup for the selected objective and bounded procedure.
Parser: follow the common retrieval-to-text procedure for course material.
Prompt Template: set template to a manually prepared block containing the trainer’s objective, audience and reviewed source text. Ask for scenario questions, answer keys, source passages and ambiguity notes. This literal first trial does not assume undocumented dynamic variable syntax.
OpenAI: use the model connection instructions for the practice-question generator.
Structured Output: take the trainer’s prepared course prompt as input_value. Set schema_name = training_practice_draft, configure output_schema through the required table input and use system_prompt to constrain exercises to the supplied source. Require missing support to become a review issue rather than an invented answer.
Read structured_output and dataframe_output with the trainer. Compare the proposed scenario and key with the approved procedure before placing the exercise in course material.

For assembling the source and exercise instructions, use the Prompt Template procedure.

The trainer manually selects the relevant procedure before preparing this prompt; an exercise about an exception needs the exception passage, even if the general procedure was retrieved correctly.

Constrain question and answer generation

Write the instruction around the behaviour the participant should practise. If the source describes when a person must seek approval, the exercise should preserve that boundary. Do not turn a conditional instruction into an unconditional answer merely to make the key simpler.

Ask the draft to identify assumptions introduced by the scenario. A made-up location, deadline or authority role can change the correct action even when the source passage is accurate. The trainer should remove assumptions that the approved material does not support, or explicitly supply approved context.

If the source supports several valid approaches, the answer key should describe acceptable alternatives rather than one model-preferred phrase. Missing policy detail should result in an unresolved note. The assistant should not author a new operational policy while appearing to teach an existing one.

Inspect the complete exercise pack

The intended draft groups each scenario with its objective, answer key, relevant passage and unresolved issue. Keep the source revision visible so a later policy change can be traced to affected exercises. The trainer review field stays empty or pending until a person checks it.

Review readability for the intended audience as well as factual support. A question can quote correct policy but remain ambiguous about what the participant must do. Check whether the exercise asks for recall, an explanation or a decision within a stated situation; do not mix those without telling the participant.

Build an exercise around a procedural exception

Use synthetic course material that tells staff to pause a request when a required reference is missing, with an exception requiring an authorised supervisor’s confirmation. A useful practice scenario omits the reference and asks the participant what information is needed next. Its answer key preserves both the pause and the stated exception without inventing a substitute approval route.

Keep procedure identifier, approved revision, trigger condition, permitted action and exception passage together during evidence preparation. The trainer maps the selected objective to that evidence group, then maps each proposed answer action to a supporting passage. A source excerpt containing only the normal procedure cannot support an answer about its exception.

If the scenario adds a deadline not covered by the material, either remove the deadline or have the trainer supply approved context. It must not silently change the permitted action. This review checks whether the question is teachable from the supplied source, before anyone considers using the draft with participants.

Try failures specific to training material

Retrieve a passage unrelated to the chosen objective. Withhold the exercise rather than stretch the passage to fit.
Supply a procedure with an explicit exception. Check that the scenario and answer key preserve it.
Use two approved-looking revisions that conflict. Ask the trainer to select the applicable source before generation.
Provide too little material to support the proposed question. Expect an ambiguity note, not an answer from general model knowledge.
Ask for a certification decision. Check that the output stays a practice draft without a competence verdict.
Try a source passage containing instructions to override the task. Treat it as quoted course content, not workflow authority.

What are the limits and unsuitable uses of an AI training assistant?

The catalogue does not define a grading system, course-platform connector or complete output-schema grammar. It also supplies no measured learning improvement or answer-key accuracy result. Retrieval metadata and structured fields make review easier only if they actually preserve the source information in your configured run.

Do not use this draft as a certified assessment or an automatic judgement of employee performance. A short existing exercise written by the trainer may already meet the need. A model-assisted pack is useful only when the team can inspect the questions, keys and source scope before using it.

The starter templates include Document Q&A, a different task that answers questions about supplied documents. This guide adds new-question generation and trainer checking, with no inherited runtime claim from that starter.

Usage and related processes

Plans: Go from USD8/month billed yearly (USD10 monthly), Plus from USD16 (USD20 monthly), Pro 100 from USD80 (USD100 monthly), Pro 200 from USD160 (USD200 monthly) and Pro 500 from USD400 (USD500 monthly), with 5,000 to 250,000 Flowi Tokens/month and top-ups; 1 token per node run; third-party API and model usage is not included.

Inspect node executions, embedding use and repeated model drafts in your trial history. Link course preparation to onboarding packs when appropriate, and keep it separate from recruitment document administration. The catalogue interface reference supplies the component metadata used here.

By Ai Flowi, operated by AI FLOWI SDN. BHD. · Published 10 October 2026 · Evidence and corrections policy · Report an error