English
EspañolArtificial intelligence software development
We build AI software for production, not demos: for scientific research, materials science and industrial automation, together with AIXC Research.
Own simulation · calculation engines with AIXC ResearchOver 100 peer-reviewed papers, at the same conferences where Google DeepMind and OpenAI publish.
AIXC Research is the Zytech Group’s AI research lab. It develops the calculation and artificial-intelligence engines we apply at Zytel.
- NeurIPS 202629 papers in workshops
- ICML · ICLR 2026AI for science and reasoning
- AAAI-26 · CIKMverifiable reasoning, finance and law
- StanfordAgents4Science 2025
- IROS 2025robotics, invited talk in Hangzhou
AI for materials science, with AIXC Research: accepted papers
- ACHT-World: Causal World Models for Closed-Loop Self-Driving LaboratoriesAI4X 2026, symposium, oral presentation
- Physics-Informed Surrogates in a Verified Agentic Pipeline for Robust Molecular SimulationML4Molecules 2025, workshop, poster
- The Architectural Immune System: A Framework for Correcting Synthetic Fallacies in AI-Driven ScienceAgents4Science 2025, symposium
David Scott Lewis
Nearly five decades in technology and an AI researcher since the 1990s. Executive roles at Microsoft, Samsung, Oracle and META Group (Gartner); accelerated MBA from Stanford and the Senior Executives programme at Harvard Kennedy School. He was Executive Vice President of Zytech Group from 2007 to 2011.
Recent papers and talks
- NeurIPS · ICML · ICLR · AAAIAIXC Research’s 100+ peer-reviewed papers, 2025–2026
- Stanford Agents4Science 2025accepted paper
- IROS 2025, Hangzhouinvited speaker, on the same programme as the founder of Unitree
Selected peer-reviewed papers
A selection of AIXC Research’s work in robotics, engineering and verifiable AI. Each title links to the official conference record.
Robotics and AI scientists
- Building a Robot Scientist: From Reasoning Models to Autonomous DiscoveryIROS 2025 · AIR4S · Hangzhou (invited talk)
- The Architectural Immune System: A Framework for Correcting Synthetic Fallacies in AI-Driven ScienceStanford · Agents4Science 2025
- ACHT-World: Causal World Models for Closed-Loop Self-Driving LaboratoriesAI4X 2026 · Singapore
- RSI for Science: A Verifier-First Framework for AI ScientistsICML 2026 · AI4Science
- Count Every Validation: Auditing Hidden Attempts in Recursive AI ScientistsNeurIPS 2026 · AI4Science
Engineering, planning and forecasting
- Does LLM Co-Evolution Pay Off in RTL Verification? A Cost-Normalised Audit on Ten RTLLM DesignsNeurIPS 2026 · AI4ChipDesign
- Recursive World-Model Self-Repair in PDDL DomainsNeurIPS 2026 · NEmo
- A Self-Improving Forecasting Agent That Chooses Its Own Validation WindowNeurIPS 2026 · FMTS
- Scientific CI/CD for Self-Modifying Discovery AgentsICML 2026 · AI4Math
Verifiable reasoning: finance and law
- VAP: Preventing Plausibility Traps in Algorithmic Trading with Physics-Informed Learning and Formal VerificationCIKM 2025 · Financial AI
- SOKRATES: Distilling Symbolic Knowledge into Option-Level Reasoning via Solver-Guided Preference OptimizationAAAI 2026 · LM Reasoning
- Glass-Box Arbitrators: An Explainable Neuro-Symbolic AI Framework for International Commercial ArbitrationAAAI 2026 · LM Reasoning
Full list of 100+ papers at aixcresearch.com/publications.
Where we apply this AI
Zytel Engineering
Our own calculation engines (DEM, finite elements and wear), developed with AIXC Research, to calculate welded parts for machinery manufacturers.
- Every figure traced to its run
- Report signed by a licensed engineer
- Silos, tanks, tippers and rotary drums
Fleet tracking
Tracking software that estimates real-time range from BMS data and proposes viable routes.
- BMS, GPS and solar input data
- Research with GITEL (I3A)
- Presented at EPoSS 2021
A full stack for AI in production
LLMs and agents
GPT, Claude, Llama, Mistral, DeepSeek. Orchestration with LangChain, LlamaIndex, MCP agents and multimodal RAG.
Applied machine learning
PyTorch, TensorFlow, scikit-learn. Time series, computer vision and forecasting.
Data and MLOps
Vector DBs (pgvector, Weaviate, Qdrant), Airflow/Prefect pipelines, Kubernetes + GPU.
On-premise and air-gapped
Local LLMs for sensitive data: llama.cpp, vLLM, Ollama, quantisation and in-house fine-tuning.
Formal verification
Symbolic analysis, model checking and invariants for autonomous agents in critical domains.
Industrial integration
Connectors to SCADA, OPC-UA, ERPs and IoT sensors. AI built into fleets and plants (Zytel EV).
In phases, with real deliverables at every stage
No endless PoCs: every phase ends with something that works and can be measured.
Discovery
We audit data, processes and the real business question. Output: technical roadmap and estimated ROI.
- Data and ETL audit
- Model and architecture selection
- KPI definition
Pilot
Working prototype on the client’s real data. Measurable value before scaling.
- MVP in pre-production
- Evaluation against baseline
- Iteration with end users
Production
Go-live, monitoring, retraining and SLA-backed support, in the cloud or on-premise.
- CI/CD and observability
- Drift detection
- Maintenance contract
An AI project in the pipeline?
Write to us and we will schedule a no-obligation discovery call.