Curriculum vitae

Ten years of experience across academia, startups, and a global enterprise, and across diverse roles as engineer, researcher, professor, architect, and now member of technical staff at an AI lab.

Experience

May 2026 – present
FirstPrinciples, Member of Technical Staff, Research Los Angeles, CA
- Research on how frontier language models and agents fail on expert-level scientific work: calibration, multi-turn robustness, error propagation, and LLM-judge reliability.
- Co-created a hybrid verification sub-system and evaluation harness combining deterministic symbolic checks (non-LLM-based), executable numerical tests, and cross-family conceptual review, with a custom ledger, translation layers, and an ablation toolkit.
- Led roadmap planning for the research unit and assumed the role of architect for non-platform delivery, including verification as a service.
Jun 2024 – Apr 2026
IBM, Senior Technical Solution Architect, Data and AI Kraków, Poland / Los Angeles, CA
- Built non-standard PoCs and MVPs for enterprise and government clients and diagnosed failing generative-AI systems. Proposed an agentic RAG system that cut hallucinations by over 80% in a client trial and became a reusable asset.
- Built an evaluation framework for LLM applications covering multi-turn use and perturbed or adversarial inputs, reporting faithfulness, robustness, consistency, and calibration separately.
- Co-developed IBM's roadmap for next-gen agentic AI in a focus group led by IBM's chief scientist; contributed research on self-adapting and self-improving agent loops.
- Technical lead of the Data and AI group: led executive client discussions, ran internal trainings and multi-day client workshops, and supervised four interns and an MSc thesis.
Jul 2020 – May 2024
University of Southern California, Visiting Assistant Professor (2023–2024), Postdoctoral Scholar (2020–2023), Visiting Researcher at USC Information Sciences Institute (2020–2024) Los Angeles, CA
- Proposed a self-supervised method for detecting microstructural transitions without labels (npj Computational Materials, first author) and built an MVP for Sandia National Laboratories.
- Research on privacy-preserving federated learning robust to data poisoning, neural emulators of quantum dynamics, and the role of context and noise in LLM-assisted generation.
- Founded and led the Machine Learning for the Natural Sciences group (30+ research students); supervised three PhD theses and eight capstone projects.
- Taught undergraduate physics and graduate machine learning. Designed the curriculum of a new graduate program that combines physics and AI.
Jul 2018 – May 2020
Fetch.ai, Lead Research Scientist (2019–2020), Machine Learning Scientist (2018–2019) Cambridge, UK
- Research on novel consensus algorithms, machine learning in decentralized multi-agent systems, and new attack vectors and architectural mitigations.
- Found a critical flaw in the main protocol three months after joining and led its redesign into a DAG-based consensus protocol (working paper, Tokenomics 2019).
- Built test suites, benchmarks, attack simulations, and PoCs. Led the development of a patented method for collective learning without sharing data (US Patent 12,749,011).
- Technical lead of a group of four. Worked alongside the head of research through a series of pivots across a research and engineering team of thirty.
Aug 2016 – Jun 2018
TypeScore, Machine Learning Engineer London, UK
- Founding engineer in a lean fintech team. Built the end-to-end ML platform from scratch (data pipelines, structured-data embeddings, NLP risk models, CI/CD, Google Cloud) and owned it in production.
- Led data collection, model training, and evaluation for the first free credit score for UK businesses.

Education and training

2020–2023
Postdoctoral Scholar. University of Southern California, Los Angeles, CA. Machine learning for the natural sciences. Department of Physics and Astronomy and the Information Sciences Institute.
2019–2020
Graduate coursework. Yale University, New Haven, CT. Theory of Deep Learning, Geometry of Data, Foundations of Machine Learning, Cryptography and Security, Economics and Computation.
2011–2016
PhD in Physics, with distinction. Jagiellonian University, Kraków, Poland. Numerical and variational methods for strongly correlated quantum many-body systems.
2012–2015
Interdisciplinary doctoral program, Society, Environment, Technology. Jagiellonian University. Competitive three-year program taught in English.
2009–2012
BSc in Computer Science. Jagiellonian University, Kraków, Poland. A second degree, taken alongside the MSc in physics.
2006–2011
MSc in Theoretical Physics. Jagiellonian University, Kraków, Poland. Including a semester at the Niels Bohr Institute, Copenhagen.

Publications

M. Abram. Emergent inference-time semantic contamination via in-context priming. arXiv:2604.04043 (2026)
M. Abram. Toward evaluation frameworks for multi-agent scientific AI systems. arXiv:2603.26718 (2026)
X. Li, H. Tang, S. Chen, Z. Wang, R. Chen, M. Abram. Why does in-context learning fail sometimes? Evaluating in-context learning on open and closed questions. arXiv:2407.02028 (2024)
L. Li, M. Abram, A. Prem, S. Haas. Hofstadter butterflies in topological insulators. Chapter in Recent Topics on Topology, IntechOpen (2024)
X. Li, H. Tang, S. Chen, Z. Wang, A. Maravi, M. Abram. Context matters: data-efficient augmentation of large language models for scientific applications. arXiv:2312.07069 (2023)
Y. Yao, C. Cao, S. Haas, M. Agarwal, D. Khanna, M. Abram. Emulating quantum dynamics with neural networks via knowledge distillation. Frontiers in Materials 9, 1060744 (2023)
M. Abram, K. Burghardt, G. Ver Steeg, A. Galstyan, R. Dingreville. Inferring topological transitions in pattern-forming processes with self-supervised learning. npj Computational Materials 8, 205 (2022)
D. Stripelis, M. Abram, J. L. Ambite. Performance weighting for robust federated learning against corrupted sources. arXiv:2205.01184 (2022)
Y. Ye, D. Qiu, J. Ward, M. Abram. Model-free real-time autonomous energy management for a residential multi-carrier energy system: a deep reinforcement learning approach. Proceedings of IJCAI 2020, pp. 339–346
M. Abram, D. Galindo, D. Honerkamp, J. Ward, J.-M. Wong. Democratising blockchain: a minimal agency consensus model. Tokenomics 2019, Paris. arXiv:2006.05390 (2020)
M. Abram, M. Zegrodnik, J. Spałek. Antiferromagnetism, charge density wave, and d-wave superconductivity in the extended t–J–U model. Journal of Physics: Condensed Matter 29, 365602 (2017)
M. Abram, M. M. Wysokiński, J. Spałek. Tricritical wings in UGe₂: a microscopic interpretation. Journal of Magnetism and Magnetic Materials 400, 27 (2016)
M. M. Wysokiński, M. Abram, J. Spałek. Criticalities in the itinerant ferromagnet UGe₂. Physical Review B 91, 081108(R) (2015)
M. M. Wysokiński, M. Abram, J. Spałek. Ferromagnetism in UGe₂: a microscopic model. Physical Review B 90, 081114(R) (2014)
M. Abram. t–t′–J–U model in mean-field approximation: coexistence of superconductivity and antiferromagnetism. Acta Physica Polonica A 126, A-25 (2014)
A. Kapanowski, M. Abram. Model of hard spheroplatelets near a hard wall. Physical Review E 89, 062503 (2014)
M. Abram, J. Kaczmarczyk, J. Jędrak, J. Spałek. d-wave superconductivity and its coexistence with antiferromagnetism in the t–J–U model: statistically consistent Gutzwiller approach. Physical Review B 88, 094502 (2013)

Patent

J. Ward, M. Abram, D. Honerkamp. Distributed computer system and method of operation thereof. US Patent 12,749,011. Filed 2021 (priority 2020), granted 2026. Application published as US 2023/0316127 A1, EP 3929742 A1, and WO 2021/260517 A1

Awards

2025
watsonx.ai Practitioner Mastery, IBM. Held by 18 people worldwide at the time.
2024–2025
Performance Award and Entrepreneur Award for Outstanding Contributions, IBM.
2021
Sponsored research award ($20,000), Sandia National Laboratories.
2020
Outstanding Reviewer, Journal of Physics: Condensed Matter. Given to about one percent of reviewers.
2013–2015
Exceptional Doctoral Performance Award, Jagiellonian University.
2006
Ninth place nationally in the Polish Astronomy Olympiad. Finalist of the Polish Physics Olympiad.

Teaching

2025
Explainable AI, lecture series, Jagiellonian University.
2024–2025
Technical workshops at IBM on information retrieval, language models and semantic search, and large-scale document processing.
2020–2024
University of Southern California: Machine Learning for Data Science (graduate), Classical Mechanics, Thermodynamics and Statistical Mechanics, Electricity and Magnetism, Electrodynamics, and Fundamentals of Physics.
2013–2015
Jagiellonian University: Statistical Physics, Physics with Elements of Biophysics, Physics Laboratory.

Service

Reviewer for Machine Learning: Science and Technology, New Journal of Physics, Journal of Physics: Condensed Matter, Journal of Physics: Materials, Superconductor Science and Technology, Physica Scripta, and IOP SciNotes.

Technical

Languages: Python, C++, Bash, R.
LLMs and agents: Large language models, agentic workflows, multi-agent systems, agent architecture, agent harnesses, tool use and function calling, Model Context Protocol (MCP), agent skills and hooks, retrieval-augmented generation (RAG), agentic RAG, context engineering, prompt engineering, in-context learning, fine-tuning and adapters, inference-time scaling, agent memory, structured outputs.
Evaluation and verification: Model evaluation, benchmark and test-suite design, evaluation harnesses, LLM-as-a-judge, cross-family judge panels, A/B testing, ablation studies, red-teaming, adversarial testing, failure-mode analysis, calibration, hallucination detection, faithfulness and grounding, citation checking, multi-turn robustness and sycophancy, symbolic verification, executable checks.
Machine learning: Deep learning, self-supervised and semi-supervised learning, knowledge distillation, curriculum learning, federated learning, inverse problems, neural emulators and surrogate models, scientific machine learning, Bayesian modeling, uncertainty quantification, explainable AI, NLP, computer vision.
Robustness and governance: Adversarial attacks, data poisoning, privacy-preserving learning, AI safety and alignment, guardrails, AI governance, model risk management.
Retrieval and data: Semantic search, hybrid retrieval (BM25, SPLADE, dense embeddings), vector databases (Milvus), Elasticsearch, MongoDB, document processing pipelines, lakehouse architecture, data governance.
Distributed systems: Decentralized protocols, consensus mechanisms, distributed ledgers, mechanism design, protocol security analysis, multi-agent coordination.
Architecture and delivery: End-to-end AI system design, solution architecture, REST services, PoC and MVP delivery, IBM watsonx (watsonx.ai, watsonx.data, watsonx.governance), technical leadership, stakeholder engagement.
Frameworks and tools: PyTorch, TensorFlow, scikit-learn, NumPy, pandas, PyMC, NLTK, LangChain, Qiskit, matplotlib, OpenAI and Anthropic APIs, Claude Code.
Infrastructure and MLOps: Docker, Podman, Kubernetes, AWS, Google Cloud, CI/CD (Jenkins, Travis CI), pytest, Google Test, Luigi, git, Linux, Slurm, high-performance computing.
Scientific computing: Numerical methods, variational methods, simulation of quantum many-body systems, quantum computing (Qiskit), LaTeX.
Practices: Agile (Scrum, XP), test-driven development, code review, mentoring.