Early-career Saarland University researcher combining LLM code generation with formal verification in AI planning
2
papers here
2
reviewed by Evolve
1
first-author
0
corresponding
—
works · OpenAlex
2
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-16) and cover the author's entire public record, not just the papers listed here. Reproducibility signal comes from Evolve's own reviews.
Research profile
Output—
An output score is not computed for this profile: the snapshot's per-year counts span only 2021–2026 and lack the venue and authorship-position detail needed to normalize production. This axis would measure total works, per-year throughput, and venue tiers.
Impact—
Citation impact is not yet scored: the only signals present are h-index 2 and i10-index 0, an early-career record too thin for stable measurement. This axis would measure citations, h-index trajectory, and influence relative to field norms.
Focus0/100
The focus axis scores 0/100, reflecting a wide topic spread rather than one dominant cluster: Logic, Reasoning, and Knowledge leads with 7 topic assignments, followed by NLP techniques (6), AI-based problem solving and planning (5), syntax/semantics/linguistic variation (4), and AI in games (2); the field distribution itself splits evenly (0.5/0.5) between AI planning and automated planning plus formal verification.
Collaboration44/100
Collaboration scores 44/100 on a compact network of 8 distinct coauthors. The strongest repeated tie is Jörg Hoffmann, listed twice among frequent coauthors; the cluster also includes Nicola J. Müller, Timo P. Gros, Naya Rudolph, Ayal Taitler, and Chaahat Jain on planning and LLM work, plus Alexander Koller on semantics — consistent with the snapshot's NLP (6) and syntax/semantics (4) topic counts. This reads as a lab-centered Saarland cluster rather than a broad international network.
Activity52/100
Activity scores 52/100. The snapshot's counts-by-year show 3 works in 2021, 1 in 2023, 1 in 2025, and a peak of 4 in 2026 — nine works total — with recent3yWorks at 5. Both Evolve-reviewed papers are 2026 arXiv releases (one first-author, one middle-author), making 2026 the most productive and visible year on record; the gaps in 2022 and 2024 hold the score near the midpoint.
Reproducibility55/100
Reproducibility scores 55/100, the mean across exactly 2 scored papers — both rated medium readiness in Evolve's review. The generalized-planning paper's Lean-kernel-checked completeness proofs (12/13 PDDL domains) and I2G's test-verified generator soundness point to machine-checked artifacts, but Evolve flags limits in both: validity constraints are user-written and the solvability check is incomplete. With only two papers scored, the mean is indicative rather than definitive.
Research summary
Katharina Stein is an AI-planning researcher at Saarland University, Germany; the profile is name-resolved from OpenAlex bibliographic data. The snapshot records nine works between 2021 and 2026 — 3 in 2021, 1 in 2023, 1 in 2025, and a peak of 4 in 2026 — with the field distribution split evenly (0.5/0.5) between AI planning (automated planning and knowledge representation) and artificial intelligence — automated planning and formal verification. Two 2026 arXiv papers are reviewed in the Evolve catalog. As first author of 'Provably Complete Generalized Planning with LLMs,' Stein couples LLM-generated generalized plans with Lean-kernel-checked completeness proofs, attaining completeness on 12 of 13 PDDL domains; as a middle author on 'Generating Instance Generators in PDDL Planning' (I2G), she contributes to LLM-generated instance generators with test-verified soundness whose efficiency strongly beats ASP and RL baselines. Topic counts place Logic, Reasoning, and Knowledge first (7), followed by NLP techniques (6), AI-based problem solving and planning (5), syntax/semantics (4), and AI in games (2). Radar scores are collaboration 44, activity 52, and reproducibility 55 (mean over 2 papers, both medium readiness); output and impact are unscored, with h-index 2 and i10-index 0 marking an early-career citation record across 8 distinct coauthors.
Collaboration pattern
The network is compact: 8 distinct coauthors in the snapshot, concentrated at Saarland University. Jörg Hoffmann appears twice in the frequent-coauthor list — the strongest repeated tie — alongside Nicola J. Müller, Timo P. Gros, Naya Rudolph, Ayal Taitler, and Chaahat Jain on planning/LLM work, plus Alexander Koller on semantics, consistent with the snapshot's NLP and syntax/semantics topic counts. Collaboration scores 44/100: a lab-centered cluster, not yet a broad distributed network.
Research keywords
Highlights
Kernel-checked completeness proofs for LLM-generated generalized plans
I2G: LLM-generated PDDL instance generators with test-verified soundness
Timeline
- 2021Earliest year in the snapshot: 3 works recorded, the joint-highest annual count before 2026.
- 20231 work recorded — a low-output year within the OpenAlex window.
- 20251 work recorded; the rolling recent-3-year total stands at 5 works.
- 2026Peak year with 4 works; includes both Evolve-reviewed arXiv papers — first-author 'Provably Complete Generalized Planning with LLMs' and middle-author 'Generating Instance Generators in PDDL Planning' (I2G).
Papers in the Evolve catalog
Frequent co-authors
About this profile
This profile is AI-structured by Evolve.Pub from public bibliographic data (OpenAlex) and Evolve's own paper reviews. It is not written or endorsed by the researcher unless the profile is claimed. Radar dimensions are computed from real bibliometrics with fixed formulas — raw figures are shown alongside; missing dimensions mean insufficient data, never zero. Errors? Claim the profile to correct or hide it.
- Bibliometrics come from an OpenAlex-resolved snapshot; 'Katharina Stein' is a common German name, so treat this as name resolution from public bibliographic data rather than a verified identity.
- h-index (2) and i10-index (0) come from the snapshot's raw record; output and impact radar scores are not computed.
- Only the 2 arXiv papers reviewed by Evolve carry qualitative verdicts; other works in the snapshot are not individually assessed.
- The reproducibility mean (55) rests on just 2 papers, both rated medium readiness, so read it as indicative.