Refactor AI panel generation to remove table_name parameter and enhance prompt clarity

[Artuomka]

No description provided.

[Copilot]

Pull request overview

Refactors AI-driven dashboard widget generation to remove the tableName query parameter and instead have the AI discover relevant tables via tool calls (getTablesList, getTableStructure), with updated prompts and E2E tests.

Changes:

• Removed tableName from the generate-widget-with-AI controller API and data structure.
• Updated the AI panel generation use case to run a streaming tool-call loop for schema discovery and improved prompt instructions.
• Updated SaaS E2E tests to use the new endpoint contract and mock tool-call streaming.

Reviewed changes

Copilot reviewed 4 out of 4 changed files in this pull request and generated 3 comments.

File	Description
backend/test/ava-tests/saas-tests/dashboard-ai-widget-e2e.test.ts	Updates E2E tests to call the endpoint without tableName and mocks tool-call streaming behavior.
backend/src/entities/visualizations/panel-position/use-cases/generate-panel-position-with-ai.use.case.ts	Replaces table-bound prompting with a tool-loop approach; adds JSON extraction and updates EXPLAIN-based refinement flow.
backend/src/entities/visualizations/panel-position/panel-position.controller.ts	Removes tableName query parameter from the API and updates Swagger docs/timeout.
backend/src/entities/visualizations/panel-position/data-structures/generate-panel-position-with-ai.ds.ts	Removes table_name from the use case input structure.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

userEmail is declared but not initialized for non-agent connections, and getUserEmailOrReturnNull() can return null. This value is then passed into the tool loop where agent DAOs require a real email for getTablesFromDB/getTableStructure. Initialize userEmail to null/undefined explicitly, widen the parameter types accordingly, and for agent connections fail fast with a clear error if the email cannot be resolved.

Suggested change

	let userEmail: string;
	if (isConnectionTypeAgent(foundConnection.type)) {
	userEmail = await this._dbContext.userRepository.getUserEmailOrReturnNull(userId);
	let userEmail: string = '';
	if (isConnectionTypeAgent(foundConnection.type)) {
	const resolvedUserEmail = await this._dbContext.userRepository.getUserEmailOrReturnNull(userId);
	if (!resolvedUserEmail) {
	throw new BadRequestException(
	'Unable to resolve user email for agent connection. A valid email is required to use AI tools.',
	);
	}
	userEmail = resolvedUserEmail;

[Copilot]

runExplainQuery calls executeRawQuery(explainQuery, '') without passing userEmail. For agent-backed connections, executeRawQuery requires both tableName and userEmail, so this will send an empty table name and an undefined email to the agent and likely fail. Pass through userEmail (and a meaningful table name if required by the agent) when calling executeRawQuery, similar to other places that execute queries against agent connections.

[Copilot]

extractJsonFromResponse slices from the first { to the end of the response, but does not trim anything after the closing }. If the model returns valid JSON followed by any trailing text, JSON.parse will still fail even after cleanAIJsonResponse. Consider extracting up to the last } (or reusing the existing JSON sanitization helper) so the parser is resilient to common LLM formatting drift.

Suggested change

	return response.slice(firstBrace);
	const lastBrace = response.lastIndexOf('}');
	if (lastBrace !== -1 && lastBrace > firstBrace) {
	return response.slice(firstBrace, lastBrace + 1).trim();
	}
	return response.slice(firstBrace).trim();