Distill

Distill

The chemistry is in the paper.
Getting it out is the work.

Distill does that work.

Reactions, chemicals and procedures — with their conditions, quantities and outcomes — read from the text, the schemes and the tables, then resolved, reconciled and balance-checked. Structured chemical data you can check.

See it on a real paper
Distill's procedure view: the paper's procedures drawn as ordered step graphs — 3-carbethoxy-7-chloro-4-hydroxyquinoline (9), 7-chloro-4-hydroxyquinoline-3-carboxylic acid (10) and 4,7-dichloroquinoline (5) in view, with the Mannich base (14) and amodiaquine dihydrochloride dihydrate (3) below — each node carrying its operation and real quantity, such as Add diethoxymethylene malonate 520 g, Heat 97–100 °C and Add POCl3 221.4 mL, with stoichiometry badges beneath the charge steps and a line carrying each intermediate into the next procedure.

The problem

The data is in the document. Getting it out takes a chemist’s time.

The reactions, the chemicals and the procedure are all there — with the quantities, conditions, equipment, yields and purities — spread across the text, the schemes and the tables of a paper, a patent or a report. Getting all of it into one structured form means going through every document, in full, every time.

That is time spent away from the chemistry. And what comes out varies with who did it, the numbers are seldom checked against each other, and once a value is in a spreadsheet nobody can say which sentence it came from.

in a paragraph

… H2O (1600 mL) was added to the flask, the heating was turned off, and the temperature was allowed to cool to 50 °C. The pH of the mixture was adjusted to 4 using a 25% aq NaOH solution. 4,7-DCQ (5, 336 g, 1.696 mol, 1.0 equiv) was added to the mixture. The mixture was then refluxed at 85 °C for 3 h, followed by stirring the mixture at 5 °C for 2 h. …

chemicalactionconditionequipment

in a scheme

Scheme 2 of the paper, 'Preparation of Amodiaquine Dihydrochloride Dihydrate (3) from 4-Acetamidophenol (13)': 4-acetamidophenol with paraformaldehyde and diethylamine to the Mannich base 14, aqueous HCl to the amine 15, then 4,7-dichloroquinoline 5 to amodiaquine dihydrochloride dihydrate 3, recrystallised from ethanol–HCl.

in a table

Table 5 of the paper, 'Reproducibility of the Developed Process at Different Reaction Scales': the Mannich base 14 at 150 g and 390 g and amodiaquine 3 at 200 g and 300 g, each run three times, with yields between 89 and 95%.
Scheme 2 and Table 5 are reproduced from Gohain et al., Org. Process Res. Dev. 2024, 28, 124–131 (doi:10.1021/acs.oprd.3c00205), CC BY 4.0. The examples on this page are Distill’s reading of that paper.

Distill reads all three automatically. Upload the document; the data comes back in minutes, in the same format every time, with the quantities cross-checked and every value linked to its sentence. The time goes back to the chemistry.

The deliverable

What you get back

Every paper comes back in the same five parts.

01

One entry per compound

The paper calls the same compound 5, 4,7-DCQ and 4,7-dichloroquinoline. Distill gives it one entry with structure, formula, CAS and InChIKey, so it is counted once and found once.

Distill's chemical registry: a table of species, each row showing the drawn structure, the compound name and id, role badges, how many reactions use it, its SMILES, formula, molecular weight, CAS number, InChIKey and a resolution status. Every row reads RESOLVED; ADQ carries four aliases and 4,7-DCQ two.
4,7-DCQ is named three ways in the paper. One row keeps all three.

02

Balanced, including the by-products

Each reaction with its reactants, reagents, solvent and products is balanced atom by atom. The by-products the paper leaves out are included, which is what you need for a mass balance and a waste estimate.

A reaction card in Distill, tagged Balanced and classified as an N-acetyl deprotection: the Mannich base 14 with water and hydrochloric acid at 85 °C for 4 h, giving the amine 15 with acetic acid as by-product. Every species is drawn and labelled with its role, chemical id and SMILES, and the balanced reaction SMILES is printed underneath.
This hydrolysis also releases a mole of acetic acid. The scheme and the procedure leave it out.

03

The procedure, step by step

Each step is labelled — charge, heat, stir, filter — with its inputs and conditions in order. Click a step to see the sentence it was read from, then open that sentence in the paper. Where the paper gives no source for a value, none is claimed.

Distill's procedure view: the five procedures drawn as ordered step graphs. One step in the amodiaquine procedure, Add 4,7-DCQ 336 g, is selected and ringed, and a panel beside it shows that step's record — the verbatim sentence it came from, '4,7-DCQ (5, 336 g, 1.696 mol, 1.0 equiv) was added to the mixture', with an 'open in document' link, the reaction it belongs to (the nucleophilic aromatic substitution giving amodiaquine, drawn), and its input 4,7-DCQ as reactant at 336 g, 1.696 mol, 1.0 equiv.
The step — its source sentence, its inputs with roles and equivalents.
The same procedure regrouped by equipment: a 'flask' row carrying the 27 steps of the final procedure, from Charge HCl (880 mL) through the washes, and a 'No equipment stated' row carrying the 54 steps of the first four procedures, for which the paper names no vessel.
The same steps, by vessel — 27 in the flask the paper names for its last procedure, 54 for which it names none. Where the paper names no vessel, the step is not assigned one.

04

What it isn’t sure about, it shows you

A consistency check runs on every extraction at no cost, with no model involved. The audit is a separate, costed step: one click, and a model reads the finished extraction and lists what looks wrong. Nothing is changed unless you accept it.

In this paper it found a published mass that does not match the moles beside it. The row is shown below, left open rather than corrected.

Distill's review panel: issues grouped by cause. The group 'Value disagrees with the paper' is open, and its first row reads 'Cited mass and moles are physically inconsistent: needs 0.6 kg, have 0.5 kg. Recorded as cited; no value substituted', anchored to four places in the document.
Findings are grouped by cause. This one is the paper’s own error, so it is left open rather than corrected. Click the row ↗

05

Export

Everything above downloads as one JSON file: every reaction, chemical, step and condition, with each value’s source sentence. Load it into a route planner, a process model or an LCA tool.

Distill's Export button, with its panel open: 'Export extraction as JSON — records in the file: Reactions 11, Steps 81, Chemicals 26, plus conditions and equipment', above a Download JSON button.
Distill's split view: the procedure spine on the left; on the right the same run as a JSON tree — Chemicals 22, Reactions 11, Procedures 5, Actions 81, Materials 64, Conditions 82 — every entry expandable, with the Export button in the toolbar above.
The same run, as your systems will read it. Every step on the left is a record on the right.

Every paper comes back in the same format. Run fifty and you can query them together.

Bring a paper.

A paper, a patent or a report: we’ll take you through what Distill makes of it.

Runs in CDI’s own AWS account. Your documents never train a model.

If you’re interested in booking a demo or would like to learn more about pricing, please contact us at info@cdi-sg.com