HomePlatformEnterpriseThe ScienceAboutCompanyInsightsPublicationsPrivacyTerms
DeepCeutix - AI Drug Design PlatformDeepCeutix - AI Drug Design Platform
Platform
Resources
Enterprise
Company
  • 01Platform
    OverviewResearch agentsBiologics agentsSafety agents
  • 02Resources
    The ScienceInsightsPublications
  • 03Enterprise
  • 04Company
    AboutPress KitContact
DeepCeutix - AI Drug Design PlatformDeepCeutix - AI Drug Design Platform

Autonomous Pharmaceutical Intelligence.
London, UK

Try the playground

Platform

  • Platform
  • Research agents
  • Biologics agents
  • Safety agents
  • Enterprise

Resources

  • The Science
  • Strategic briefings
  • Publications

Company

  • About
  • Contact
  • Press Kit

Trust

  • Trust Centre
  • Privacy
  • Terms
All Systems Operational
© 2026 DeepCeutix Ltd. // Engineered in London
© 2026 NVIDIA, the NVIDIA logo are trademarks and/or registered trademarks of NVIDIA Corporation in the U.S. and other countries.

Research suite

Seven research
specialists.

Formulation, optimization, FTO, patent drafting, VCM process design, and analytics, each agent anchored to IID precedent, Drugs@FDA, EMA EPAR, and the global pharma patent corpus. Outputs every formulation chemist actually uses: candidate cards, delta memos, FTO maps, USPTO-formatted drafts.

Fig 01 · Research workflow

Seven agents.
One choreography.

Research informs Formulation. FTO clears chemistry before scale-up. Optimization refines the candidate. VCM sets the manufacturing window. Patent Drafting captures the work. Analytics ranks the features that move outcomes and turns those rankings into design constraints for the next cycle.

The threading bar walks the full candidate-development flow in order, from Research through to Analytics as the seventh and final step.

research-agent1RESEARCHAGENTfto-agent2FTOformulation-agent3FORMULATIONoptimization-agent4OPTIMIZATIONvcm-agent5VCMPROCESSpatent-drafting-agent6PATENTDRAFTINGanalytics-agent7ANALYTICSAGENT

Per agent

Each specialist
in detail.

Click any agent in the choreography above to jump to its deep section.

01Research

Formulation Agent

Designs the formulation. Cites the IID precedent.

  • →QTPP, CQAs/CMAs/CPPs, ICH Q9 risk assessment, control strategy.
  • →3-5 candidates from distinct strategic angles, each with rationale and IID precedent.
  • →Each card carries components, process, properties, QC specs, and safety profile.
  • →Dispatches to VCM or Optimization when API physicochemistry demands it.

“Design an oral formulation for this poorly soluble compound.”

Example prompt

Reads the indication, the API physicochemistry, and the program target product profile. Proposes formulations grounded in IID excipient ceilings, GRAS conclusions, and the precedent dosage forms in Drugs@FDA. Operates per ICH Q8(R2). When the API physicochemistry suggests a specialized form (BCS Class II → ASD, BCS Class IV → cyclodextrin or lipid carrier, sensitive biologic → lyophilized cake), the agent dispatches to VCM Process Design or Optimization to refine the manufacturing window before returning the final candidate.

Returns: 3-5 candidate cards (F01-F0N) with rationale, IID precedent, and a one-line risk note on each.

F02Oral, modified-release tablet
Recommended

MR Matrix (HPMC K100M)

Composition
ExcipientFunction% w/w
ActiveAPI35
HPMC K100MHydrophilic matrix former30
Lactose monohydrateFiller25
Microcrystalline celluloseFiller / binder8.5
Magnesium stearateLubricant1
Colloidal silicon dioxideGlidant0.5
BCS classII
Target Tmax6-8 h
Target stability24 mo at 25°C / 60% RH
ManufacturingDirect compression
IID precedentNDA 022030Mirapex ER 0.375 mg
Rationale

HPMC K100M selected as the primary release-control matrix at 30% w/w; concentration sits within IID precedent for oral MR products on Mirapex ER (NDA 022030). Lactose monohydrate and MCC chosen for direct-compression compatibility and to stabilise the API in a low-water-activity matrix.

Fig. Candidate card returned by the Formulation Agent (illustrative).

02Research

Formulation Optimization Agent

Refines a working candidate against a single objective.

  • →Cost, bioavailability, stability, manufacturing-efficiency, or regulatory-alignment objectives.
  • →Variants anchored to parent (F02 → F02A, F02B…) so trade-offs sit side-by-side.
  • →Substitutions cross-checked against 40k+ FDA Drug Master File listings.
  • →Flags every substitution that voids a regulatory precedent.

“Optimize this tablet for 20% cost reduction while maintaining dissolution.”

Example prompt

Takes a working candidate and pushes it against a single objective: cost, manufacturability, dissolution, or stability. Substitutes excipients within IID-precedented ranges. Cross-references substitutions against the FDA Drug Master File listings (40k+) to confirm a qualified supplier exists at scale. Flags every substitution that would void a regulatory precedent.

Returns: A revised candidate card plus a delta memo against the original.

RETURNSVariant +deltaF02 → F02A/B/Cbefore-after · tradeoffs
Fig. Formulation Optimization Agent deliverable.
03Research

Research Agent

Synthesizes the literature with primary-source citations.

  • →PubMed (37M records), ClinicalTrials.gov + AACT, EMA EPAR, Drugs@FDA reviews.
  • →Comparative analysis across formulation strategies; conflicts flagged explicitly.
  • →Technology landscape surveys, weighted to 2020-present for state-of-the-art questions.
  • →Optional knowledge-graph traversal to surface non-obvious connections.

“Compare lipid-based vs amorphous solid dispersion approaches for BCS Class II compounds.”

Example prompt

Compares formulation strategies across PubMed (37M records), ClinicalTrials.gov + AACT, EMA EPAR, and Drugs@FDA reviews. Cites primary sources with DOI, PMID, or application number. Flags conflicting evidence. Returns a single coherent comparative memo rather than a paragraph of prose with citations sprinkled in.

Returns: A comparative memo with a tier-anchored citation list.

RETURNSComparativememoDOI · PMIDapplication number
Fig. Research Agent deliverable.
04Research

FTO Agent

Scores patent risk before chemistry. Treats claims as design constraints.

  • →0-100 risk scoring banded Low / Low-Moderate / Moderate / Moderate-High / High.
  • →Blocking patent identification with verbatim claim language.
  • →Six design-around strategies: excipient, concentration, combination, process, dosage form, route.
  • →Orange Book analysis with Drug Substance / Drug Product / Use Patent classification.

“Analyze FTO for an HPMC-based extended-release tablet for Drug X.”

Example prompt

Risk-scores each candidate scaffold (0-100) against the global pharma patent corpus, identifies blocking claims line by line, and proposes design-arounds. For ANDA programs, the agent reads the Orange Book, distinguishes Drug Substance / Drug Product / Use Patents, and recommends a Paragraph III versus Paragraph IV filing strategy.

Returns: An FTO map with claim-by-claim blocking analysis and three non-infringing alternatives.

RETURNSFTO mapclaim-by-claim3 alternativesOrange Book
Fig. FTO Agent deliverable.
05Research

Patent Drafting Agent

Drafts USPTO-formatted patent applications.

  • →Title, Field, Description of Related Art, Summary, Detailed Description, Definitions, tables.
  • →10-20 numbered claims with 2-3 independents (composition, method-of-treatment, manufacturing).
  • →4 Examples including a required Comparative Example for non-obviousness argumentation.
  • →Explicit [PLACEHOLDER] / [REQUIRES ICH STABILITY STUDY] markers, never fabricates experimental data.

“Draft a USPTO nonprovisional for F02 with three independent claims.”

Example prompt

Writes the specification, dependent claims, and prior-art distinction memo against MPEP §608 and 37 CFR §1.84. Pulls embodiments from the candidate cards already produced earlier in the conversation. Drafts (PA01, PA02…) stream live into the side panel as the agent writes. The agent does not fabricate experimental data: it uses explicit [PLACEHOLDER - TO BE DETERMINED], [REQUIRES ICH STABILITY STUDY], and [REQUIRES DISSOLUTION STUDY] markers for stability, dissolution, comparative-study, and bioequivalence data, since those have to come from your laboratory work.

Returns: A USPTO-formatted draft application; your patent counsel signs.

RETURNSUSPTO draftspec · 10-20 claimsIDS-ready
Fig. Patent Drafting Agent deliverable.
06Research

VCM Process Design Agent

Recommends manufacturing parameters for VCM, HME, spray drying.

  • →VCM screens drug-polymer pairs in minutes per formulation, vs 30-60 min for HME, near-zero material loss.
  • →Polymer Tg + 20-50°C, below degradation onset (typically 160-230°C, 4-10 min).
  • →Material-sparing protocols: 0.5-2 g total, 8/20/25 mm disc geometries.
  • →Working knowledge of Parteck MXP, Kollidon VA 64, Soluplus, Eudragit EPO, HPMCAS, Klucel, Methocel.

“Design a VCM screen for ITZ on Kollidon VA 64 vs HPMCAS-L.”

Example prompt

Specialized for MeltPrep Vacuum Compression Molding, with HME pre-screening built in. Sets process windows for VCM, hot-melt extrusion, and spray drying. VCM screens drug-polymer pairs in minutes per formulation, versus 30-60 minutes for HME, with near-zero material loss. Working knowledge of common ASD polymers (Parteck MXP, Kollidon VA 64, Soluplus, Eudragit EPO, HPMCAS-L/M/H, Klucel, Methocel).

Returns: A VCM process card with operating ranges and the boundary studies needed to qualify them.

RETURNSVCM processcardVCM01-N · rangesboundary studies
Fig. VCM Process Design Agent deliverable.
07Research

Analytics Agent

Builds the model from your data. Feeds insights back to other agents.

  • →Random Forest, gradient boosting, logistic regression on uploaded experimental datasets.
  • →Six-step loop: explore, plan, ask clarifications, execute, visualize, write.
  • →Publication-ready figures + PDFs with LaTeX typography; exports as MD/LaTeX/CSV.
  • →Feeds top-ranked features back to Formulation and Optimization as new design constraints.

“Run a DoE-matrix analysis on this dissolution dataset and report optimal parameter ranges.”

Example prompt

Trains predictive models on uploaded experimental datasets (Random Forest, gradient boosting, logistic regression). Ranks the features that move the outcome and feeds those features back to the Formulation and Optimization agents as new design constraints. The sandbox runs server-side, so results stream identically to web, desktop, and mobile.

Returns: A research-grade PDF report with methodology, results, and feature-importance rankings; updated constraints applied to the conversation context.

RETURNSResearch-gradePDFmethodology · figuresfeature ranks
Fig. Analytics Agent deliverable.

Biologics suite

Three biologics specialists

Biologics Research, FTO, and Formulation & CMC, anchored to ICH Q5A-E, Q6B, Q13, Q14, FDA Purple Book, EMA biosimilar register, SAbDab, and PLAbDab.

Safety suite

Six safety specialists

ERA, extractables, leachables, nitrosamines, OEL, and PDE, anchored to ICH M7, EMA HBEL, USP <1664>, and ECETOC TR 101.

Architecture

How the platform routes

Reality Anchor, agent routing, foundation, regulatory frameworks, and data sources: the unified platform overview.

Book a demo.

Tell us what you are working on (molecule, indication, stage, open question).

A formulation lead will be in touch within a business day.

By submitting, you agree to our Privacy Policy.


info@deepceutix.com|London, UK|LinkedIn