Three pillars. One coherent practice.
Whether you need an expert second opinion, a system shipped end-to-end, or your team levelled up — we structure the engagement around the outcome you need, not a fixed menu.
Expert Consultancy
Deep, hands-on engagement on the AI problems where the cost of being wrong is high and the value of getting it right is enormous.
Time-series forecasting
Univariate and multivariate forecasting with MRN ensembles, gradient-boosted models, and probabilistic neural nets — for finance, macroeconomics, demand planning, and operations.
Classification & decision systems
Calibrated, explainable classifiers for risk, churn, conversion, fraud, sports outcomes, and other binary/multi-class decisions — with SHAP (SHapley Additive exPlanations) feature attribution, local (per-prediction) and global (per-dataset), built in as standard.
Model explainability & SHAP analysis
Explainability is built into delivery, not added afterwards. For tree-based and tabular models, SHAP quantifies the contribution of each feature to individual decisions (local) and across the entire dataset (global); for neural forecasting models we use gradient-based attribution (integrated gradients). Stakeholders see exactly why a model decides what it decides, with auditable, defensible attributions.
Statistical inference & causal evaluation
Hypothesis testing, confidence intervals, causal identification, and A/B test evaluation — turning interventions into measurable, defensible business decisions, with counterfactual analysis and treatment/control experimental design.
Algorithmic & systematic trading
Strategy research, signal generation, walk-forward validation, execution wiring, and risk-managed live deployment.
GenAI architecture review
Independent assessment of LLM, RAG, and agentic AI initiatives — from prototype to production-readiness — delivered in partnership with our team of expert contractors.
AI strategy & roadmaps
Boardroom-grade advice on where AI investment will earn its return — and where it will not.
Collaborative Systems Development
We embed with your team to build production AI systems together — versioned, monitored, CI/CD-deployed — so the capability stays with you long after we leave.
End-to-end ML pipelines
From data ingestion through feature engineering, modelling, deployment, and monitoring — built as one coherent system, not handed-off artefacts.
Cloud deployment
AWS (EC2, EBS, RDS), Cloudflare Pages and Workers, containerised services, and infrastructure-as-code where it pays back.
MLOps & CI/CD
GitHub Actions, MLflow, automated testing, and reproducible builds. If a model can't be retrained on a button press, we haven't finished.
Real-time data integration
Live wiring into Finage, OANDA, MetaTrader, FRED, internal warehouses, and bespoke APIs — with clean failure modes and replay.
Monitoring & retraining
Drift detection, alerting, and scheduled retraining built in from sprint one — so models stay calibrated as the world changes.
Inspiring Workforce Education
Tailored, evidence-based training delivered by an active researcher and 20+ year practitioner with a proven Russell Group teaching record.
Executive AI literacy
Half-day to multi-day workshops that give leadership teams the vocabulary, mental models, and decision frameworks for the AI era — without hype.
Hands-on technical bootcamps
Code-along training for analysts and engineers in Python, scikit-learn, deep learning, time-series ML, and the modern GenAI stack — GenAI modules delivered in partnership with our team of expert contractors.
University-grade curriculum design
Drawing on the Nottingham Trent MSc Data Analytics for Business programme that Jon designed and led — adaptable to your in-house academy.
Doctoral & research supervision
6 PhDs supervised to completion. Available for industrial doctoral co-supervision and KTP-style partnerships.
Specialist seminars
Topical deep-dives on MRN ensembles, time-series validation, and AI risk management, plus RAG architectures and agentic AI delivered in partnership with our team of expert contractors.
Frequently asked questions
What does Perceptronix do?
Perceptronix Ltd is the UK-based AI consultancy of Dr Jonathan A. Tepper, founded in 2018. It builds production-grade, explainable machine-learning systems for time-series forecasting, classification and decision support, and delivers AI training for executives and engineers.
Who leads the work?
Dr Jonathan A. Tepper, PhD, FHEA, founder and lead data scientist. Twenty-five years in neural networks and machine learning, spanning eighteen years as an academic and eight running production AI systems commercially. He supervises MSc projects at Aston Business School, guest-lectures on the University of Birmingham's MSc FinTech, and is co-investigator and technical lead on the EPSRC-funded MASCET project with the University of Birmingham and NIESR. For work outside his own specialisms, Perceptronix draws on a small team of expert contractors in the US, India and Dubai, working under his direction.
How do you make models explainable?
Explainability is built into delivery rather than added afterwards. For tree-based and tabular models we use SHAP to attribute each prediction to its features, locally and across the dataset; for neural forecasting models we use gradient-based attribution (integrated gradients) and, in our research, knowledge extraction from the network's internal state. In every case the aim is the same: stakeholders can see why a model decided what it decided, in terms they can audit.
What kinds of forecasting problems do you take on?
Univariate and multivariate time-series forecasting for finance, macroeconomics, demand planning and operations. For example, the MASCET inflation model built with NIESR and the University of Birmingham is typically within ±0.2 percentage points of the ONS monthly CPI outturn.
Do you use GenAI and LLMs?
Where they earn their place. Our core expertise is time-series forecasting, classification and explainable ML. For LLM, RAG and agentic-workflow builds we work with a small team of expert contractors in the US, India and Dubai, under Jon's technical direction, so we can scope, deliver and quality-assure GenAI work. We also use LLMs in our research to make forecasting models interpretable to non-specialists.
Do you only advise, or do you also build?
Both. We can provide an expert second opinion, or embed with your team to build end-to-end ML pipelines with MLOps, CI/CD, monitoring and retraining, with a team of expert contractors where the build needs more hands than one, so the capability stays with you.
Where are you based, and do you work internationally?
Perceptronix is based in the UK (Derbyshire) and works with clients globally, supported by contractors in the US, India and Dubai.
How do engagements start?
With a confidential, no-obligation first conversation that usually ends with a clear view of feasibility, data needs and time-to-value. Get in touch via the contact page or jtepper@perceptronix.net.
Have a forecasting or classification problem that needs to work in production?
Tell us about it. First conversations are confidential, no-obligation, and usually end with a clear view of feasibility, data needs, and time-to-value.