# Chris von Csefalvay

> AI researcher and computational epidemiologist. Principal at HCLTech's AI Practice, leading post-training research and clinical intelligence. Author of two technical books. Based in Denver, CO.

## Who He Is

Chris von Csefalvay is a Principal at HCLTech's AI Practice, leading post-training research and clinical intelligence. He specialises in post-training for language models, agentic AI systems, and computational epidemiology, with nearly two decades of experience in data science and AI across enterprise, life sciences, and public health.

He is the author of:
- *The Craft of Post-Training: A Practical Guide for AI Engineers and Developers* (No Starch Press, available July 2026) — a comprehensive guide to SFT, RLHF, GRPO, DPO, evaluation discipline, LoRA/QLoRA, and domain adaptation for enterprise AI practitioners
- *Computational Modeling of Infectious Disease* (Elsevier, 2023) — a monograph on agent-based modelling, disease dynamics, and computational epidemiology

He holds degrees from the University of Oxford and Cardiff University. He is a Fellow of the Royal Society for Public Health and a Senior Member of IEEE. ORCID: 0000-0003-3131-0864.

## Areas of Expertise

- Post-training for language models: supervised fine-tuning (SFT), reinforcement learning from verifiable rewards (RLVR), GRPO, DPO, KTO, preference optimisation
- Agentic AI systems: multi-agent frameworks, verifiable agentic architectures, agent interaction protocols, agentic safety
- Large language models: evaluation, LLMOps, model routing, efficiency (LoRA, QLoRA, quantisation, distillation)
- Computational epidemiology: agent-based modelling, epidemic dynamics, geospatial epidemiology
- AI governance and ethics
- Pharmacovigilance and biomedical AI

## Key Writing (The Notebook)

### Post-training and reinforcement learning
- https://chrisvoncsefalvay.com/posts/post-training-instrument-cluster/ — The post-training instrument cluster, Part I: SFT and evaluation
- https://chrisvoncsefalvay.com/posts/post-training-instrument-cluster-rl/ — The post-training instrument cluster, Part II: reinforcement learning
- https://chrisvoncsefalvay.com/posts/post-training-instrument-cluster-grpo/ — The post-training instrument cluster, Part III: GRPO
- https://chrisvoncsefalvay.com/posts/post-training-scale/ — Post-training: three disciplines in a trenchcoat
- https://chrisvoncsefalvay.com/posts/deepseek-seatbelts-and-straitjackets/ — Seatbelts and straitjackets: DeepSeek and AI safety

### Agentic AI and multi-agent systems
- https://chrisvoncsefalvay.com/posts/two-years-of-agents/ — What I talk about when I talk about agentic AI
- https://chrisvoncsefalvay.com/posts/skeuomorphic-fallacy/ — The skeuomorphic fallacy in agent design
- https://chrisvoncsefalvay.com/posts/context-engineering/ — The Snowmobile Symptom: context engineering
- https://chrisvoncsefalvay.com/posts/programmatic-agentic-ai/ — Why we need to move from prompting to programming
- https://chrisvoncsefalvay.com/posts/after-agents/ — After agents: what comes next
- https://chrisvoncsefalvay.com/posts/agents-agora/ — After agents, part 2: Agents and the Agora
- https://chrisvoncsefalvay.com/posts/team-of-rivals/ — Teams of Rivals: multi-agent coordination
- https://chrisvoncsefalvay.com/posts/choral-ai/ — From agents to the Chorus
- https://chrisvoncsefalvay.com/posts/dorkestration/ — Dorkestration: agent orchestration patterns
- https://chrisvoncsefalvay.com/posts/mcp-mcpmark/ — A slow walk out of Dikika Cave: MCP and MCPMark
- https://chrisvoncsefalvay.com/posts/agentic-simulation/ — Between the motion and the act: agentic simulation
- https://chrisvoncsefalvay.com/posts/agentic-speciation/ — Deep kimchi: agentic speciation

### AI philosophy, governance and predictions
- https://chrisvoncsefalvay.com/posts/war-for-the-soul-of-ai/ — The War for the Soul of AI
- https://chrisvoncsefalvay.com/posts/tlon/ — The world will be Tlön
- https://chrisvoncsefalvay.com/posts/five-wild-guesses-2026/ — Five wild guesses for 2026
- https://chrisvoncsefalvay.com/posts/five-wild-guesses/ — Five unconventional predictions
- https://chrisvoncsefalvay.com/posts/moral-maps/ — The Moral Pulse of the Machine
- https://chrisvoncsefalvay.com/posts/ai-opportunities-action-plan/ — Deja Vu, All Over Again: AI policy
- https://chrisvoncsefalvay.com/posts/hype-slop-craft/ — The Hype, the Slop and the Craft

### LLMs: language, hallucinations and evaluation
- https://chrisvoncsefalvay.com/posts/llms-language/ — Beyond Broca: LLMs and language
- https://chrisvoncsefalvay.com/posts/asemantic-induction-of-hallucinations/ — Asemantic induction of hallucinations in LLMs
- https://chrisvoncsefalvay.com/posts/knowledge-dividend-of-llms/ — The knowledge dividend of large language models
- https://chrisvoncsefalvay.com/posts/soya/ — SOYA: the only benchmark that matters
- https://chrisvoncsefalvay.com/posts/prompt-engineering/ — Prompt Engineering: The Art of Yesterday
- https://chrisvoncsefalvay.com/posts/noise-in-the-neurons/ — Just noise in the neurons
- https://chrisvoncsefalvay.com/posts/lair/ — LAIR: Language As Intermediate Representation
- https://chrisvoncsefalvay.com/posts/lyre-of-hephaestus/ — The Lyre of Hephaestus
- https://chrisvoncsefalvay.com/posts/metadynamic-ai/ — The end of isotropy and the rise of metadynamic AI
- https://chrisvoncsefalvay.com/posts/dwarf-fortress/ — The hardest AI problem you've never heard of
- https://chrisvoncsefalvay.com/posts/academic-generative-ai/ — Stochastic parrots, cap and gown edition

### Epidemiology and public health
- https://chrisvoncsefalvay.com/posts/data-for-the-next-pandemic/ — Data for the next pandemic
- https://chrisvoncsefalvay.com/posts/mrna-gene-therapy/ — Why the COVID-19 vaccines are not gene therapy
- https://chrisvoncsefalvay.com/posts/paxlovid/ — Peace, love, Paxlovid and Pfizermectin
- https://chrisvoncsefalvay.com/posts/95-percent-myth/ — The 95% myth

### Other topics
- https://chrisvoncsefalvay.com/posts/ai-girlfriends/ — Love in the Time of Algorithms
- https://chrisvoncsefalvay.com/posts/ai-human-interaction/ — What I learned from getting bodied by a robot
- https://chrisvoncsefalvay.com/posts/biome-engineering/ — Biome engineering
- https://chrisvoncsefalvay.com/posts/ransomware-prisoners-dilemma/ — Ransomware attacks as n-person prisoner's dilemmas
- https://chrisvoncsefalvay.com/posts/as-we-may-see/ — As we may see: the world after dashboards
- https://chrisvoncsefalvay.com/posts/julia-a-post-mortem/ — Julia: a post-mortem

## Publications

- *The Craft of Post-Training: A Practical Guide for AI Engineers and Developers* (No Starch Press, available July 2026): https://posttraining.guide
- *Computational Modeling of Infectious Disease* (Elsevier, 2023): https://computationalinfectiousdisease.com
- DAEDRA: A language model for predicting outcomes in pharmacovigilance (2024): https://arxiv.org/abs/2402.10951
- Full publication list: https://chrisvoncsefalvay.com/papers/

## Teaching

Visiting thesis supervisor for the data science track at the Budapest University of Technology and Economics. Details: https://chrisvoncsefalvay.com/teaching/

## External Profiles

- GitHub: https://github.com/chrisvoncsefalvay
- Google Scholar: https://scholar.google.com/citations?user=X_2G-VsAAAAJ
- ORCID: https://orcid.org/0000-0003-3131-0864
- Wikidata: https://www.wikidata.org/wiki/Q107095298
- LinkedIn: https://www.linkedin.com/in/chrisvoncsefalvay/