We work on problems where the method matters — molecular design, conformational sampling, simulation, clinical modelling and cryptographic infrastructure — and we run them as research engagements, not software projects.
Every service below is delivered by the same research group. That means the person who designs the method also runs the computation, interrogates the result, and writes it up — and every engagement ends with something you can reproduce, not a black box.
Each area is described in plain language first, with the methods named alongside it.
We train conditional generative models — variational autoencoders paired with U-Net samplers and structure-guided decoders — on the sequence–structure–function relationships of a family of known actives, then sample new candidates under explicit design constraints: target geometry, binding-site chemistry, stability and tractability for synthesis.
Candidates are ranked by predicted affinity and filtered in silico, so bench time is spent on a short, defensible list rather than on a screen.
In plain terms: we teach a model what a good answer looks like, then ask it to propose new ones — hundreds at a time, ranked — before anyone commits to the lab.
Flexible peptides and intrinsically disordered regions do not have a single native structure, so random sampling wastes effort revisiting similar shapes while missing whole regions of conformational space. Our Digital Nets Conformational Sampling (DNCS) method represents a peptide by its backbone dihedral angles and draws well-spread, low-discrepancy samples of them using digital nets, then scores every conformation with standard force fields (AMBER99SB, AMBER-FB15) and ranks by Boltzmann weighting.
DNCS is published in Physical Chemistry Chemical Physics, released as open source (DNCS 1.0, Rust, OpenMM-compatible), and can be applied to your sequences or adapted to your system.
In plain terms: instead of wandering a huge building hoping to find every room, we walk a planned route that is built to cover every wing.
We set up, run and analyse all-atom molecular dynamics of proteins, peptides and ligand–biomolecule systems, and — more importantly — we connect the trajectories to a question: which interactions persist, which conformations dominate, how does a mutation, a binder or a biomarker change the picture.
This is the work behind our collaboration with Balaji Dental College, where measurable biomarkers are linked to the molecular behaviour underlying dental disease.
In plain terms: we film how the molecules actually move, then analyse the footage to explain why a system behaves the way it does.
We build supervised models on imaging and structured data with the evaluation discipline clinical use demands: leakage-free splits, calibrated probabilities, quantified uncertainty, and external validation before any claim of performance. The aim is a consistent second read alongside a clinician’s own judgement, never a replacement for it.
This is the approach we apply in our research collaboration with RELA Hospitals on AI-based liver tumour prediction from medical imaging.
In plain terms: a carefully tested second opinion that tells you how sure it is, and when it is not.
We work in the representations that make chemical and biological data computable — molecular fingerprints and graphs, sequence and structure descriptors — and use them for similarity search, property prediction, library triage and annotation. This is the groundwork that lets generative design and simulation start from a well-posed problem.
Where a partner has a large compound or sequence collection, we reduce it to a shortlist that is worth the cost of making and testing.
In plain terms: turning a warehouse of molecules or sequences into a short list worth opening.
Encryption that is safe today can be recorded now and broken later. We assess where a platform’s long-lived data is exposed, design a staged migration to NIST-standardised post-quantum algorithms with hybrid key exchange, and build in crypto-agility so the next change is a configuration update, not a rewrite.
Our deeper note on the threat model and migration path is on the Quantum Security page and in our Insights.
In plain terms: protect the data so it stays protected even when computers get much more powerful.
Research that cannot be rerun is not research. We package methods as reproducible, documented pipelines — containerised environments, version-controlled workflows, and databases for experimental and clinical data — on Unix and HPC systems, and we release tools openly where the partner agrees, as we did with DNCS.
This is what turns a successful pilot into something your own team can operate.
In plain terms: so the result is something you can run again tomorrow, and hand to the next person.
Most partners begin small and let the evidence decide what comes next.
A short, bounded assessment: we audit your data and the prior art, define the quantitative success criteria, run a baseline, and give you a written go / no-go recommendation with the evidence behind it.
We develop and validate a method against your problem, produce a result set you can act on, and hand over the code and the report. This is where most collaborations show whether they are worth scaling.
For problems that need several iterations of method, computation and experiment: a shared roadmap, regular scientific review, co-authored publications and open-source releases where appropriate, formalised through an MoU as with RELA Hospitals and Balaji Dental College.
Method disclosure, intellectual property and publication terms are agreed at the start of an engagement, and your data stays under your governance throughout.
Tell us what you are trying to design, model or secure, and we will propose the smallest engagement that can answer it.
Start a feasibility study