Chemistry & Chemicals
Chemical Structure Recognition & Correction
Recognition and drafting correction for full compound names, Markush, indefinite structures and complex substituents.
OCSRstructure drafting
Expert Data Services · Annual-Report Edition
Domain experts and AI working together to deliver traceable industry data
Focused on chemistry, new materials, pharma R&D and biological sequences. Schema co-design, AI pre-annotation, expert refinement and three-tier QC — every record traces back to its source text and annotator.
Chemistry & Chemicals
Chemical Structure Recognition & Correction
Recognition and drafting correction for full compound names, Markush, indefinite structures and complex substituents.
Pharma R&D
Bioactivity Data Structuring
Structured extraction of test substances, activity metrics, value units and assay methods.
New Materials
New-Material Patent Data Processing
Structuring of solid-state electrolyte formulations, process steps, characterization data and performance metrics.
Pharma R&D
Biomedical Patent Sequence-Relation Annotation
Annotation of antibody sequence–antigen relations, heavy/light chain pairing and binding affinity constants.
§ 01
Delivery Pipeline
Four steps, one thread · logged end to endDelivery PipelinePIPELINE
STEP 01
Schema co-design
Define fields, types, granularity and boundaries with the algorithm team
STEP 02
AI pre-annotation
The model produces a first draft; experts revise it
STEP 03
Expert refinement
Experts with chemistry / materials / pharma / biology backgrounds rewrite every record
STEP 04
Three-tier QC
Agent checks, team review, expert arbitration
Human–AI effort splitEFFORT SPLIT
AI pre-annotationFirst draft (coverage varies by project)
Expert refinementLine-by-line review
QC releaseExpert sampling / full review
Data shape
chemistry.v2
Compounds / Structures / Relations
EntityN-(4-hydroxyphenyl)acetamide
SynonymParacetamol
TypeSmall-molecule drug
StructureSMILES / InChI / structure image
Full compound name
Markush
Indefinite structures
Structure correction
§ 03
Quality System
Full three-tier QC — no unreviewed data is releasedThree-tier QCTHREE-TIER QC
LEVEL 1
Automated checks by agent + knowledge base
Automated validation of field completeness, unit consistency and relation conflicts against the schema and the domain knowledge base
Full pre-check
LEVEL 2
Review by the data QC team
Dedicated QC specialists sample at the agreed ratio or review in full, flag doubtful records and send them back for correction
Coverage as agreed
LEVEL 3
Arbitration by domain experts
Experts with chemistry / pharma / materials backgrounds rule on disputed records, and conclusions are written back into the annotation guideline
Final arbitration
Three-tier
QC mechanism
Record-by-record
source and annotator traceable
§ 04
Data Compliance
Six clauses · written into the contractSix data compliance clausesSIX CLAUSES
Clause 01
NDA signed
Confidentiality agreement and authorized-personnel list completed before kick-off.
Clause 02
Data stays in your domain
Processed inside the client-designated environment or isolated domain.
Clause 03
No secondary use
Never reused for other clients, never used to train our own models.
Clause 04
Isolated deployment available
Supports dedicated cloud environments, client VPCs or on-premise deployment.
Clause 05
Access control & audit trail
Role-based permissions with exportable audit logs.
Clause 06
Deletion after delivery
Raw and intermediate data deleted on the agreed schedule.
§ 05
Case Studies
Real delivery scenarios
Pharma R&D
Bioactivity Data Structuring
Structured extraction of test substances, activity metrics, value units, test subjects and methods.
PK/PDtable extraction
Biological Sequences
Biological Sequence Extraction
Extraction and normalization of protein/nucleic-acid sequences, modification details, sequence numbering and cross-page information.
sequence normalizationmodification decomposition
New Materials
New-Material Patent Data Processing
Structuring of solid-state electrolyte basics, process steps, characterization data and performance metrics.
solid-state electrolytepatent processing
Biological Sequences
TCM Knowledge Base Construction
Cleaning of herb, component, target and disease data and construction of a searchable candidate-component library.
knowledge basemulti-table cleaning
Pharma R&D
Biomedical Patent Sequence-Relation Annotation
Annotation of antibody sequence–antigen relations, heavy/light chain pairing and binding affinity constants KD/IC50/EC50.
antibody sequencesaffinity
Pharma R&D
Pharma Pipeline Data Updates
Continuous monitoring of drug type, target/MOA, indication, R&D stage and pipeline status.
pipeline monitoringDMP
New Materials
Material Patent Entity & Property Annotation
Annotation of material entities, elements/components, weight percentages, performance properties and manufacturing processes.
material entitiesproperty annotationTypical deliverablesDELIVERABLES
- 01Structured datasetJSON / CSV / Excel, fields mapped one-to-one to the schema
- 02Annotation guidelineField definitions, edge cases, acceptance criteria
- 03QC reportSampling ratio, error taxonomy, rework log
- 04Traceability logEach record maps to its source location, annotator and revision time
§ 06
Domain Data Notes
On data in vertical AI · updated regularlyNo industry news, no product pitches. "Domain Data Notes" writes about one thing only: when AI lands in a specific industry, where exactly the data layer gets stuck. It covers scenario difficulties, engineering methods, compliance requirements and trend judgement.
Why Annotation Accuracy Is Misleading
EngineeringAnnotation accuracy measures the samples that were labelled, not model behaviour. Drawing on public research, this article explains why a dataset of positives alone cannot define a model's boundary, why annotator disagreement is signal rather than noise, why evaluation sets saturate with age, and what each of the three PEN data categories — Positive, Edge, Negative — solves.
2026-10-01
§ 07