User rules
Separate fixed conditions from preferences. A budget cap can disqualify an offer, whereas price weighting affects the ranking of offers that met the conditions.
BotCentralHub Lab
The Lab is BotCentralHub’s working space for designing AI systems and preparing their validation. We are currently developing the personal agent concept, whose recommendations stem from user rules.
Neutral Personal Agent
Concept · Model demonstration
The design aims to create a personal agent that compares options based on user conditions and allows reviewing why a particular option was recommended. The user specifies priorities, strict limits, and actions requiring approval.
The published form consists of a four-layer design proposal and an illustrative calculation below. The model table is neither the output of a running AI agent nor proof of its neutrality.
Four layers of the working model
Comparison using identical rules
Apply identical criteria and visible weights to all offers. Commercial ties or offer sources must not covertly alter user priorities.
Decision audit log
Retain used inputs, rules, calculations, and disqualification reasons. For real offers, documenting the source and timestamp of retrieved data will also be required.
Human approval
Display recommendations prior to any potential action. The user can adjust priorities, reject the output, or approve a specific next step.
Model matrix: price and delivery speed
Fictional offers and rules created for this demonstration. These are not current prices, actual deals, or purchasing recommendations.
We compare three offers for identical equipment. The stated total price includes shipping. Other attributes are considered equal for the purpose of this demonstration.
Fixed conditions
- Total price at most 2 000 CZK.
- Delivery no later than 5 days from order.
An offer that fails to meet any fixed condition does not participate in scoring. A missing price or delivery timeframe must be supplied prior to evaluation.
Scoring eligible offers
A higher point value indicates a better result according to the chosen rule. Point scales are intentionally simple and serve solely for this demonstration.
Price scoring:
- Up to 1 600 CZK inclusive: 5 points.
- Over 1 600 CZK up to 1 800 CZK inclusive: 4 points.
- Over 1 800 CZK up to 2 000 CZK inclusive: 3 points.
Delivery scoring:
- Up to 2 days inclusive: 5 points.
- Over 2 days up to 4 days inclusive: 3 points.
- Over 4 days up to 5 days inclusive: 1 point.
By default, price has a weight of 60% and delivery speed 40%.
Total score = price points × 0.6 + delivery points × 0.4
| Offer | Total price | Delivery | Score | Result |
|---|---|---|---|---|
| A | 1 800 CZK | 2 days | 4.4 | Highest score |
| B | 1 600 CZK | 4 days | 4.2 | Meets conditions |
| C | 1 400 CZK | 7 days | — | Disqualified: delivery exceeds 5 days |
Why offer A is recommended
A receives 4 points for price and 5 for delivery: 4 × 0.6 + 5 × 0.4 = 4.4. B receives 5 points for price and 3 for delivery: 5 × 0.6 + 3 × 0.4 = 4.2. Offer C is cheaper, but exceeds the fixed delivery limit.
What changes with different priorities
If the user increases price weight to 80% and reduces delivery weight to 20%, A receives a score of 4.2 and B a score of 4.6. The recommendation changes to B. Fixed conditions remain identical, so C is not evaluated further.
The demonstration explains the impact of rules and priorities on the outcome. The calculation can be performed by standard software; it does not require AI in itself. Any potential role of a model in data retrieval or interpretation must be validated separately.
What the first prototype should validate
The following steps represent a validation plan. They are not a list of completed tests.
-
Input ingestion
Accept offers, fixed conditions, and weights. Detect missing or invalid data and request completion.
-
Repeatable calculation
Translate model rules into an isolated calculation module and compare the outcome against specified values.
-
Edge cases
Verify limit breaches, missing data, tie scores, and priority changes. On a tie, declare a tie result or apply an additional user-specified rule.
-
Recommendation review
Present selected offers, disqualification reasons, and applied rules to a human. Initial validation concludes with a recommendation without executing a purchase.
Open design questions
- How to retrieve comparable and up-to-date offer data?
- How to highlight limited source selection or commercial ties?
- How to verify that scoring reflects actual user priorities?
- What to recheck if price or availability changes prior to approval?
These questions guide further work on the design. The model table does not address them yet.
Ecosystem connection
ThinkAsimov provides space for questions of trust and the human relationship with AI. RobotYellowPages may in the future assist with navigating tools and providers. Any eventual catalog integration into this concept is a proposal for future development.
Explore related workflows
On the AI Agents page, you will find guidance for choosing the right architecture. Automation shows how to connect individual steps with approval and error handling.