3 flagged before month end Overview
Twelve fictional pharmacies. Three are flagged before month end, and eight are on track.
Case study · Client work · Resorsi
Pharmacy Rebate Engine
Rebate and pricing engine for a pharmacy client. Rules do the math, an LLM only explains it.
Overview
It reads the daily wholesaler files, matches each drug to its contract, and shows every pharmacy where it stands on its rebate tier and what it takes to reach the next one. The engine recommends the next purchasing move.
- 01 Reads the daily wholesaler files and matches each drug to its contract
- 02 Deterministic rules compute every number, the LLM only explains them
- 03 Every output cites its source rows and the rule version
- 04 Recommendations, not orders: it does not place orders
Wide social cut · 20s · with music
See the math
A shorter cut of the same purchasing walkthrough: fictional pharmacies, one tier gap, a labelled example.
Muted · unmute in the controls
The video uses fictional pharmacies. Recommendations are the product’s built-in synthetic example, labelled on screen. It is not a recording of a client’s purchasing data.
How it works · real screenshots
Tier math, traced
00 /06
Six screens from the walkthrough
Scroll to lift each one out.
From the portfolio overview to the net effect of a labelled synthetic example.
01 · Screenshot
Overview
Twelve fictional pharmacies. Three are flagged before month end, and eight are on track.
3 flagged before month end
02 · Screenshot
Pharmacies
The same fictional portfolio as a list.
Rebate position
03 · Screenshot
One pharmacy
One fictional pharmacy, with the gap to the next tier.
The gap to the next tier
04 · Screenshot
The rule steps
The stored rule steps behind that gap. The number is an example from the fictional file.
Stored rule steps
05 · Screenshot
Recommendations
Recommendations for a labelled synthetic example, not for the twelve pharmacies above.
Labelled synthetic example
06 · Screenshot
Net effect
The same example, priced as money saved against rebate given up. These are sample figures.
Saved vs. rebate given up
3 flagged before month end 
Rebate position 
The gap to the next tier 
Stored rule steps 
Labelled synthetic example 
Saved vs. rebate given up How it works · real screenshots · 01—06 3 flagged before month end 01 / 06 Rebate position 02 / 06 The gap to the next tier 03 / 06 Stored rule steps 04 / 06 Labelled synthetic example 05 / 06 Saved vs. rebate given up 06 / 06
Tier math, traced
3 flagged before month end Rebate position Pharmacies
The same fictional portfolio as a list.
The gap to the next tier One pharmacy
One fictional pharmacy, with the gap to the next tier.
Stored rule steps The rule steps
The stored rule steps behind that gap. The number is an example from the fictional file.
Labelled synthetic example Recommendations
Recommendations for a labelled synthetic example, not for the twelve pharmacies above.
Saved vs. rebate given up Net effect
The same example, priced as money saved against rebate given up. These are sample figures.
Results · fictional walkthrough data
Every number, traced to its source.
12 fictional pharmacies in the walkthrough
3 flagged before month end
8 on track
The engine shows each pharmacy its rebate tier and the gap to the next one, with every number traced to its source. Built at Resorsi. It runs in a staging environment, not in production. The figures above come from the fictional walkthrough, not a client result. No public app URL.
Full walkthrough · 52s · no audio
Every step, uncut
Overview, one pharmacy’s breakdown, and the labelled recommendations example. Captions are on screen. No narration or music.
Built at Resorsi
My role.
01
My role
Forward Deployed AI Engineer at Resorsi. Built the pharmacy rebate engine: file ingestion, drug-to-contract matching, and the tier compliance and net price logic.
02
Math you can audit
Rebate money cannot depend on a model’s guess. Rules compute tier compliance and net price, and the LLM only explains the result in plain words. Every output points back to its source rows and rule version.
03
About the demo
This walkthrough uses fictional pharmacies. It is not a recording of Halo purchasing records. The recommendations screen is the built-in synthetic example, and it is labelled as one.
Next case study · 05 of 08
Community Intelligence