Today
Lab cabinets hold notebooks from three decades and card boxes full of recipe cards, almost all handwritten. They record which raw materials were tried, what flocculated, what foamed and what was too expensive. Because nobody can search them, the lab plans trials it ran long ago, and the knowledge retires with every chemist.
An example
Orvelon is developing a version of the plant cleaner AR 240 without EDTA because a dairy asks for it. The agent searches 212 scanned lab notebooks from 1991 to 2016 and 1,800 recipe cards and finds 46 trials with gluconate or MGDA as complexing agent. Among them is a 1998 series that flocculated after three weeks at 40 °C and a 2004 series that stayed stable but was dropped with the note “too expensive”. It maps the old trade names to today's raw materials in the ERP via CAS numbers. The lab manager starts from the 2004 formulation instead of from scratch and saves two trial series.
Your data, wherever it lives
Your lab knowledge sits today in handwritten lab notebooks and card boxes of recipe cards, in trial reports as PDFs on network drives, in the LIMS and in raw material data in the ERP. demi reads, understands and links every trial with raw materials, conditions and result, with your rights. The AI finds in seconds what has already been tried, and your lab manager decides where the next trial starts.
Where the data lives today
- Scans & paper
- PDFs
- Folders & drives
- Databases
- ERP
How demi solves it
Three layers in this order: connect the data first, then the rights, then the AI.
- 1 · ConnectData Layer
Data sources in this case: scanned lab notebooks and recipe cards, trial reports as PDFs on drives, the LIMS with analyses and stability tests, and the ERP with raw materials, formulations and safety data sheets. demi replaces nothing and turns every trial into a record: raw materials with quantities and lots, order of addition, temperature, stirring time, measured values and result. Old trade names are mapped to today's raw materials via CAS number and data sheet.
- 2 · PermitPermission Layer
Formulations are core knowledge: notebooks and cards are visible only to development and the lab manager, formulations for customers under a non-disclosure agreement only to that customer's team. The lab confirms whether a transferred trial counts as checked, and no formulation goes to production or a customer without the lab manager's release. Every access is logged.
- 3 · ActIntelligence Layer
The agent reads handwriting, tables and margin notes, converts quantities into shares of the total mass and marks uncertain digits for review. To questions such as “Which trials used gluconate instead of EDTA, and how stable were they?” it answers with a list of trials, each with page and cut-out from the notebook. If a measured value is missing, the gap stays in the record instead of being estimated.
How it runs
- 1
Lab notebooks and recipe cards are scanned once; drives, LIMS and ERP are connected.
- 2
demi reads every page and transfers raw materials, quantities, conditions and results into tables, with a reference to the page.
- 3
Old trade names are mapped to today's raw materials in the ERP via CAS number and data sheet.
- 4
For a new development the lab asks in plain words and gets every similar trial with its result.
- 5
The lab manager chooses the starting point, and new trials go straight into the same data.
What changes
No trial is repeated just because it sits on paper
New formulations start from the best earlier result
Decades of lab knowledge stay when experienced chemists leave
What does it bring you?
Check it with your own numbers
Every starting value is an assumption you can change. The calculation sits right below, step by step.
Your numbers
Result
+108 hfreed per year
Assumptions and calculation
demi finds earlier trials with page and result; you choose the starting point.
2 h → 1.4 h per case
How we calculate
15 trial searches × 2 h × 30% × 1.0 (manual) × 12 months = 108 h
Value per year, before demi costs
Freed per year+108 h
Assumption: labour cost of €49.50 per hour in German manufacturing in 2025 (Destatis), plus workplace, IT and overheads.
= Value per year, before demi costs
+€7,020
gained per year
108 h ÷ 1,760 h per role = 0.1 full-time roles
An estimate, not a promise. The starting values are assumptions for an industrial company with 150 to 500 people.
Combine with more cases and get the packageQuestions about this use case
Can demi read handwriting from old lab notebooks?
Yes, from scans, including tables, crossings-out and margin notes. The agent marks uncertain digits and abbreviations and shows the cut-out so someone from the lab can confirm them. Confirmed values are marked as checked.
How does demi find raw materials that no longer exist under that name?
Via CAS numbers, old safety data sheets and supplier documents. The agent assigns each old trade name to its substance and shows which current raw material in the ERP matches it. Without a reliable match the name stays flagged instead of being guessed.
Who may see the formulations?
Only those cleared for it, usually development and the lab manager. Formulations for customers under a non-disclosure agreement are visible only to that customer's team, and every access is logged. A formulation goes to production or customers only after release.
Do new trials go in as well?
Yes. Whether in the LIMS, in a spreadsheet or still in a paper notebook that is scanned regularly: new trials land in the same data and are searchable right away, next to the old ones.
