AI Retrosynthesis · Building-Block Aware

Plan synthetic routes
from target to purchasable

An in-house retrosynthesis platform: a fine-tuned language model seeds routes, then a recursive search decomposes any molecule down to real, buyable Enamine building blocks — with drug-property scoring, reaction conditions, and literature precedent built in.

⚗️ Launch Planner Explore modules

238k

Enamine building blocks

91+

Reaction templates

2M

Training reactions

12

R&D modules

What's inside

Synthesis Planner

Predict, verify, and reconstruct retrosynthetic routes with step-by-step reaction verification.

Run multiple sampling modes and models, then rank routes by quality, purchasability, and brevity.

Process whole compound libraries from CSV in the background, with per-target results.

Drug Properties

Lipinski, Veber, Ghose, QED and a real Ertl synthetic-accessibility score for any molecule.

Analog Generator

Expand hits via catalog similarity search, scored and ranked against the parent molecule.

Scrape PubChem, ChEMBL, Wikipedia and PubMed into a searchable RAG knowledge base.

How a route is found

1

Seed

The fine-tuned SynLlama model proposes candidate disconnections for the target.

2

Search

A recursive engine applies curated retro-templates, decomposing each fragment further.

3

Match

Every leaf is checked against 238k Enamine building blocks by Morgan-fingerprint similarity.

4

Verify

Each step is round-trip verified, scored, and annotated with reaction conditions.