RESEARCHER · SCIENTIST · INNOVATOR · ENTREPRENEUR

From Scientific Discovery to Real-World Impact.

Where molecular science meets intelligent technology.

I'm Prem Kumar Perumal, a scientist working at the intersection of molecular microbiology, aerosol science and emerging technologies. I transform scientific evidence into practical solutions — advancing AI evaluation, threat modelling and biosecurity through research, innovation and real-world application.

An entrepreneur with a passion for global marketing and branding.

MOLECULAR MICROBIOLOGY / AEROSOL SCIENCE / AI & BIOSECURITY

Explore My Research Discover My Projects
Professional portrait of Prem Kumar Perumal in a blue suit and red tie
Prem Kumar PerumalMolecular microbiology · Aerosol science · AI evaluation

MY PERSONAL PHILOSOPHY

“Never let anyone define the limits of your imagination. Discover what others believe is impossible.”
— Prem Kumar Perumal
MOLECULAR DIAGNOSTICS✦EXPERIMENTAL RESEARCH✦MODEL EVALUATION✦SCIENTIFIC LEADERSHIP

SCIENTIFIC FOUNDATION · RESEARCH · INNOVATION

Science Beyond Discovery.
Evidence Beyond Assumptions.

Understanding how things work is only the beginning. Challenging what we think we know is where innovation begins.

My scientific journey bridges molecular microbiology, infectious disease diagnostics and aerosol science, combining research experience at UKHSA Porton Down and its predecessor organisations with doctoral research at the University of Bristol.

From investigating tuberculosis biomarkers and evaluating molecular diagnostic assays to studying bacterial biofilm decontamination, my work has focused on one fundamental question: How do we turn scientific evidence into reliable, meaningful outcomes?

My research in aerosol science extends this approach to the behaviour, transport and fate of airborne particles within indoor environments, exploring how physical processes influence exposure and inform our understanding of airborne risks.

Today, I am extending this scientific foundation into artificial intelligence, machine learning evaluation, threat modelling and biosecurity — bringing the same rigour used in experimental science to emerging technologies.

For me, innovation is not simply about what technology can achieve. It is about what the evidence demonstrates, what remains uncertain and how we can responsibly push the boundaries of what is possible.

UKHSA Porton Down
Molecular microbiologyUniversity of Bristol
Doctoral research in aerosol science

02 / APPLIED AI PORTFOLIO

From principles
to working tools.

Interactive projects that make methods inspectable
and results open to scrutiny.

DEMO VERSION · For a detailed project discussion or a full version, contact [email protected].
BIOEVALEVIDENCE WORKBENCH
Every answer deserves
an evidence check.
  1. Compare scientific answers with references
  2. Capture two-reviewer judgements
  3. Train an error-risk classifier
  4. Benchmark and export the evidence
EXPERT REVIEW × MACHINE LEARNING
INTERACTIVE APPLICATION · ML-ASSISTED REVIEW

BioEval Evidence Workbench

A review workflow for scientific AI answers.

Import model outputs, compare them with supplied references and record expert judgements. Train a transparent classifier to prioritise answers for review.

  • Blind review, annotations and inter-rater agreement
  • Question-group separation and held-out testing
  • Compare logistic regression with a lexical baseline
  • Threat modelling, safeguard evidence and report exports
Launch BioEval ↗

Try the threat-modelling demo ↗

Sourced biosurveillance study and reproducibility pack ↗

Includes constructed demo answers. Textual triage is not independent factual verification; real-model claims require qualified review and independently sourced evidence.

PROTEIN EVIDENCE EXPLORER

Explore a structure.
Inspect the contacts.

Published coordinates · Interactive 3D view · Geometric contact analysis

STRUCTURAL BIOLOGY · DEMO

Protein Evidence Explorer

Explore human carbonic anhydrase I with acetazolamide or methazolamide using two published protein–ligand structures.

  • Rotate the backbone and inspect bound ligand atoms
  • Adjust the contact cutoff and highlight nearby residues
  • Export atom-pair distances and source provenance
Launch protein–ligand demo ↗

Published-structure geometry, not predictions of new binders, binding affinity or biological activity. No ML or docking is run.

◈ EVIDENCE LABMODEL VALIDATION
Evidence before
confidence.
TRAIN → VALIDATE → TEST Illustrative preview
INTERACTIVE APPLICATION · VERSION 1

Evidence Lab

Surveillance data quality. Practical ML evaluation.

A browser-based workbench that trains a logistic regression classifier on synthetic surveillance-data quality records, then separates threshold selection from held-out testing.

  • Reproducible data generation and training-only preprocessing
  • Sensitivity, specificity, ROC AUC and confidence intervals
  • Calibration, error review and locked acceptance criteria
  • External model-score CSV evaluation and report export
Launch the application ↗
METHODOLOGY NOTEASSAY LOGIC / BIO EVALS
What a diagnostic lab
knows about
evaluating AI.
  1. Define the measurand
  2. Specify the comparator
  3. Set acceptance criteria
  4. Report both error directions
  5. Establish reproducibility
  6. Preserve an audit trail
MEASUREMENT · EVIDENCE · JUDGEMENT
METHODOLOGY NOTE · PERSONAL WORKING ARGUMENT

Assay Logic for Bio Evals

What a diagnostic lab knows about evaluating AI.

A framework connecting clinical laboratory validation with the evaluation of frontier AI biological capabilities. The note explores how to define the question, specify a baseline and make evaluation findings reviewable.

  • Six transfers from laboratory method validation
  • A conceptual uplift and refusal matrix
  • A proposed validation-report structure
  • Limits of the analogy and the role of expert judgement
Read the methodology note ↗

Conceptual methodology; no live model testing or empirical evaluation results. Personal views, not institutional guidance.

AEROEVIDENCESYNTHETIC SENSOR DEMO
Airborne signals.
Interpretable screening.
  1. Particle number and size
  2. Fluorescence and humidity context
  3. Trained logistic regression
  4. Error analysis and held-out testing
AEROSOL SCIENCE × ML × BIOSECURITY
INTERACTIVE APPLICATION · SYNTHETIC DATA

AeroEvidence

Bioaerosol screening through an aerosol scientist’s lens.

A simulated environmental monitoring workbench that combines aerosol features with a trained classifier to flag biological-like patterns for review.

  • Five scenarios, including dust and fluorescent interference
  • Live simulated traces and size distributions
  • Inspect model contributions and false positives
  • Lock a threshold, reveal test results and export evidence
Launch the bioaerosol demo ↗

Research demonstration using synthetic data. No pathogen identification, viability determination or field validation.

A portfolio built around evidence.Projects are described by what they implement, the data they use and the limitations of their results.

03 / SELECTED PUBLICATIONS

Research you
can trace.

Prem Kumar Perumal beside his indoor aerosol research poster at EAC 2025

RESEARCH IN PRACTICE / EAC 2025

Following particles.
Understanding exposure.

Presenting research on indoor aerosols, particle size distributions and the impact of ventilation in a naturally ventilated townhouse.

Why I chose aerosol science

The questions dramatised in the film Outbreak—how infectious threats spread and how science can help contain them—connect closely with my scientific interests. During the COVID-19 pandemic, these questions took on immediate practical importance. While contributing to diagnostic evaluation work, I became increasingly interested in gaps in understanding aerosol transmission and indoor exposure. This motivated my doctoral research in aerosol science at the University of Bristol, connecting microbiology with particle behaviour, dispersion and ventilation. The work has wider relevance to understanding bioaerosol contamination and evaluating measures to reduce exposure in built environments.

“Characterisation of Indoor Aerosols from salt atomisation and pan frying: size distribution and ventilation impact in a naturally ventilated townhouse”

European Aerosol Conference 2025 · University of Bristol research

Selected outputs. Journal articles, conference proceedings and related supporting outputs are identified separately. Follow each DOI for the publication record and full author list.

04 / LEADERSHIP & PUBLIC CONTRIBUTION

Bringing people together.
Making knowledge useful.

01

Scientific convening

Director · Graviton Global Holdings Ltd

Supporting international conferences in pharmacovigilance, vaccines and public health, connecting scientific discussion with professional exchange.

My entrepreneurial interests combine a passion for global marketing and branding with bringing people, ideas and organisations together across international audiences.

02

Community contribution

Volunteer Organiser · EVAWG Wiltshire Summit 2026

Helped deliver the Ending Violence Against Women and Girls Summit, attended by Her Majesty The Queen. Worked alongside Lady Lansdowne and the organising team, supporting press coordination and arrangements relating to royal protocol.

Explore the coalition’s work ↗
Lady Lansdowne, Her Majesty The Queen, Dr Monica Devendran and Prem Kumar at Marlborough College; caption identifies the EVAWG Summit organisers
At the EVAWG Wiltshire Summit, Marlborough College, 23 October 2026.
Photograph: Graviton International. Original image caption retained.
Prem Kumar Perumal at the lectern bearing the EVAWG Summit 2026 sign
At the lectern during the EVAWG Summit 2026.
Supporting the delivery of the summit.

05 / QUALIFICATIONS & DEVELOPMENT

A foundation in
experimental science.

Doctoral research in Aerosol Science

University of Bristol

Research on indoor aerosol dispersion and exposure in built environments.

MSc Applied Biomolecular Technology

University of Nottingham

Developing Solutions Scholar.

BTech Biotechnology

SRM University, India

First Class.

CONTINUING THE CONVERSATION

Explore the methods.
Examine the evidence.

Start an experiment in Evidence Lab ↗