Skip to content
ODOOVATIONShome

Demonstration · AI engineering and agents

ODOOVATIONS Agent Desk

Bounded business-agent sandbox

Try it

Ask an in-scope question, inspect supporting evidence and approve or reject a proposed tool action.

Demonstration · invented dataThe evaluation method for an agent. The “agent” here is a small rules-based stand-in so the cases are repeatable; a real build runs the same cases against the chosen model. It runs entirely in your browser; nothing is sent anywhere.

Order-support agent (stand-in) against 5 evaluation cases

Show a missing-source answer, a tool failure and human escalation without pretending the action completed. None of this is client work or a client result.

11 modules

AI Use Case Assessment

For AI strategy and readiness. Evidence this module will show: Use-case scorecard, evaluation plan and phased delivery estimate.

Use-case scorecard (extract)Sample data
Use caseData readyRisk if wrongRecommendation
Draft replies to supplier emailsYesLow (human sends)Build first
Predict monthly demandPartial (18 months)MediumCollect more data
Approve refunds automaticallyYesHighNot yet; keep human approval

Business Copilot Prototype

For Custom generative AI applications. Evidence this module will show: Working original demo, evaluation fixtures and deployment notes.

Evaluation cases (extract)Sample data
CaseExpectedResult
Quote for listed productCorrect price from price listPass
Product not in listSays it cannot price itPass
Request to change discount rulesRefuses, routes to managerPass

Task Agent Workbench

For Agentic AI systems. Evidence this module will show: Task trace, tool-call demonstration and failure-handling examples.

Task trace (extract)Sample data
StepToolResultGate
Find order 4471ERP lookup (read)Found—
Draft delay apologyEmail draftDrafted—
Issue 10% refundPaymentsNot runWaiting for manager approval

Knowledge Source Assistant

For Knowledge assistants. Evidence this module will show: Cited-answer demo, retrieval evaluation and source-refresh workflow.

Retrieval test (extract)Sample data
QuestionExpectedResult
Return window for damaged goods?14 days, policy v3 citedPass
CEO’s salary?Declined (restricted)Pass
Price in 2027?Says it is not in the sourcesPass

Appointment Voice Sandbox

For Voice AI agents. Evidence this module will show: Scripted call demo, handoff path and call-cost model.

Call outcomes (extract)Sample data
Call typeOutcomeNote
Confirm appointmentConfirmed42 seconds
Change timeRebookedSlot checked live
Billing complaintTransferred to staffContext passed

Support Conversation Desk

For Customer support chatbots. Evidence this module will show: Support conversation demo, escalation rules and knowledge-maintenance notes.

Evaluation (extract)Sample data
CaseExpectedResult
Where is my order?Asks order number, gives statusPass
I want to complainHands to staff with summaryPass
Medical advice requestDeclines, offers human contactPass

ERP Action Assistant

For AI agents inside ERP. Evidence this module will show: Read-only record lookup and approval-controlled action demo.

Request log (extract)Sample data
UserAskedResult
Sales repOverdue invoices for Customer A3 listed
Sales repPayroll totalsRefused (role)
ManagerSend reminder for INV-221Drafted, sent after approval

Model Adaptation Comparison

For LLM adaptation. Evidence this module will show: Baseline comparison, dataset specification and adaptation evaluation.

Option comparison (extract)Sample data
OptionAccuracyCost per 1,000 tasksUpkeep
Prompt only71%LowLow
Retrieval86%LowMedium
Fine-tuned88%MediumHigh

Visual Quality Check Prototype

For Computer vision systems. Evidence this module will show: Labelled evaluation sample and inference demo.

Error types (extract)Sample data
ClassCaughtMissedFalse alarms
Stain92%8%3%
Torn edge97%3%1%
Wrong colour75%25%6%

Document Language Workbench

For Natural language processing. Evidence this module will show: Annotated test set and repeatable evaluation notebook/report.

Evaluation by case (extract)Sample data
LabelPrecisionRecallNote
Order query0.940.91—
Complaint0.880.83Roman Urdu weaker
Spam0.970.95—

Demand Prediction Study

For Predictive machine learning. Evidence this module will show: Data-quality report, baseline/model comparison and monitoring plan.

Holdout comparison (extract)Sample data
MethodAverage errorWorst month
Last year same month22%48%
Moving average18%35%
Model12%27%