LLM post-training and token-efficient reasoning researcher; co-author on Rufus-Air and last author on CounterRoute
2
papers here
0
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
—
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-09-28) 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—
No reliable publication-volume data is available in the current record, so output remains unscored. This axis would measure total paper count, venue types, and annual production rate across the full publication history.
Impact—
Impact cannot be scored yet: no citation counts, h-index, or venue-level influence metrics are in scope. This axis would measure citations, citing works, and downstream uptake of the two reviewed papers.
Focus—
Focus is unscored because topic-classification coverage is insufficient. This axis would measure how concentrated the research agenda is; the two reviewed papers both sit in LLM post-training and efficient reasoning, but the sample is too small to score.
Collaboration63/100
This is the only dimension with enough data to score: 63/100, computed from 23 distinct coauthors in the public bibliographic record, twelve of whom recur across projects — including Ruochen Jiao, Besnik Fetahu, Zhenyu Shi, Xiaotian Han, and Hongye Jin. Both cataloged papers are multi-author team efforts, with Nigam in a middle position on Rufus-Air and last author on CounterRoute, consistent with steady but not yet dominant collaborative leadership.
Activity—
Activity is unscored: the record contains only two 2026 arXiv entries, with no multi-year publication cadence. This axis would measure publishing frequency, recency, and continuity of output over time.
Reproducibility—
A reproducibility score is not yet computed from bibliometric evidence. As qualitative context, Evolve rated both reviewed papers medium for reproducibility readiness. This axis would measure code, data, and artifact availability across the full body of work.
Research summary
Priyanka Nigam is an early-stage machine-learning researcher working at the intersection of large language model (LLM) post-training and inference-efficient reasoning. The Evolve catalog currently holds two 2026 arXiv papers. The first, Rufus-Air, is an open, eight-stage post-training recipe built on GLM-4.5-Air-Base; Evolve's review reports that it beats the official GLM-4.5-Air release and praises its unusual candor, while cautioning that the rationale behind the stage ordering is untested. The second, CounterRoute, proposes counterfactual reinforcement learning with hierarchical credit assignment so that Qwen3 models self-route between deliberate thinking and direct answering, cutting token usage by 41–51% while lifting accuracy; the review flags single-seed experiments and author-run baselines. Both papers carry a medium reproducibility-readiness rating. Nigam appears in a middle author position on Rufus-Air and as last author on CounterRoute, suggesting growing leadership of the reasoning-efficiency line. The collaborative footprint is comparatively well documented: 23 distinct coauthors and a collaboration dimension score of 63 out of 100, with twelve recurring collaborators including Ruochen Jiao, Besnik Fetahu, and Zhenyu Shi. Because no verified affiliation, ORCID, or citation record is in scope, this should be read as an early-stage profile built from public bibliographic data and Evolve's paper reviews.
Collaboration pattern
Collaboration is the best-documented axis: 23 distinct coauthors (collaboration score 63/100), with twelve recurring names — Ruochen Jiao, Besnik Fetahu, Zhenyu Shi, Chia-Yuan Chang, Renyuan Cheng, Rui Feng, Xiaotian Han, Yuan He, Hongye Jin, Linwei Li, Shiyang Li, and Fenglin Liu. Both cataloged papers are sizeable team efforts, with Nigam in a middle slot on Rufus-Air and last author on CounterRoute, suggesting a move from contributor toward leading the efficient-reasoning line.
Research keywords
Highlights
An open 8-stage post-training recipe reported to beat the official GLM-4.5-Air release
Counterfactual RL lets Qwen3 self-route reasoning, cutting tokens 41–51%
Timeline
- 2026Publishes Rufus-Air on arXiv: an open, eight-stage post-training recipe on GLM-4.5-Air-Base reported by Evolve to beat the official GLM-4.5-Air release.
- 2026Publishes CounterRoute on arXiv, introducing counterfactual reinforcement learning with hierarchical credit assignment for self-routed reasoning in Qwen3 models.
- 2026Both 2026 outputs land in LLM post-training and inference-efficient reasoning, with 23 distinct coauthors on record — the basis of the current collaboration score of 63/100.
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.
- Bibliometric signals (coauthor count, collaboration score) come from public bibliographic data (OpenAlex-style sources); citation and venue coverage may be incomplete.
- No verified affiliation, country, or ORCID is available; 'Priyanka Nigam' is a relatively common name, so this profile is name-resolved from public bibliographic records and may mix homonymous authors.
- Scope is limited to two Evolve-reviewed 2026 arXiv papers; no peer-reviewed venue record or citation data is included.
- Output, impact, focus, activity, and reproducibility dimensions lack sufficient underlying data and are deliberately left unscored.