Early-career first-author researcher in efficient LLM inference, quantization, and persistent-memory agent reliability
3
papers here
0
reviewed by Evolve
3
first-author
0
corresponding
—
works · OpenAlex
0
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-17) 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—
Output is unscored: the underlying OpenAlex record reports h-index 0, i10-index 0, and indexes works only for 2026, which is too thin to characterize productivity. This axis would measure cumulative publication volume and consistency over time.
Impact—
Impact is unscored: citation counts, h-index, and citing works are not yet available in the underlying data, and the reviewed papers are 2026 arXiv preprints too recent for citation accumulation. This axis would measure scholarly uptake and influence.
Focus—
Focus is unscored because the indexed topic distribution is too sparse to characterize concentration. Preliminary OpenAlex topics lean toward topic modeling and NLP techniques, but this axis would measure topical concentration versus breadth once more works are labeled.
Collaboration14/100
The collaboration axis scores 14 out of 100, and the raw record explains why: OpenAlex counts just 1 distinct co-author across the indexed works, with Shekar Ramachandran the single recurring collaborator in Evolve's catalog. All three reviewed papers are first-authored arXiv preprints without a corresponding-author role. This is a compact, early two-person working pattern rather than a broad network, so collaborative reach should be read as limited but focused.
Activity52/100
The activity axis scores 52 out of 100. The raw record shows 5 works in the last three years, all falling in 2026 in countsByYear, including the three arXiv preprints reviewed by Evolve. That density signals a strong current production burst, but with only one indexed year there is no baseline for judging sustained output, so the mid-range score reflects recency without longitudinal depth.
Reproducibility—
Reproducibility is unscored because no systematic artifact data, such as code releases, datasets, or replication reports, is available in the underlying record. This axis would measure whether published results can be independently reproduced; Evolve's reviews rate the three preprints' repro readiness from medium to high, which alone cannot yield an axis score.
Research summary
Jundong Hu is an early-career researcher working at the intersection of efficient large language model inference and LLM agent reliability, publishing as first author on arXiv preprints. Evolve's catalog includes three 2026 first-authored papers: XMerge on cross-axis selection and reconstructive layer merging for LLM depth compression, The Memory Trust Gap on capability-dependent failures in persistent-memory agents, and The Structure of Quantization Damage in LLMs, a causal study of post-training quantization. Across these, a consistent agenda emerges: making LLMs cheaper to deploy through depth compression and quantization, while rigorously characterizing where compressed and memory-augmented models fail. Evolve's verdicts emphasize methodological care: a well-controlled causal analysis in the quantization study, rigorous benchmark design in the memory-trust work, and broad seven-model evaluation in XMerge, tempered by notes on single-seed runs and evidence limited to 0.6-8B models on synthetic tasks. The public bibliometric record, name-resolved from OpenAlex, is young: h-index and i10-index are 0, five works are indexed for 2026, and one distinct co-author recurs in the catalog. Only collaboration (14) and activity (52) are scorable radar dimensions; output, impact, focus, and reproducibility remain unscored for insufficient data. This is an early-stage profile built from a sparse public record plus Evolve's paper reviews.
Collaboration pattern
The record shows a compact collaboration footprint: OpenAlex counts 1 distinct co-author, and Evolve's catalog surfaces Shekar Ramachandran as the recurring collaborator. All three reviewed 2026 papers are first-authored arXiv preprints without a corresponding-author role, suggesting a small, stable two-person working mode centered on efficient-LLM and agent-reliability topics rather than a large multi-group network.
Research keywords
Highlights
XMerge: select-then-refit layer merging for LLM depth compression
The Memory Trust Gap: stale-memory over-trust is capability-gated
Quantization damage is diffuse, and global g128 granularity beats oracle local repair
Timeline
- 2026Published XMerge as first author on arXiv, proposing cross-axis selection and reconstructive layer merging for LLM depth compression, evaluated across seven models.
- 2026Published The Memory Trust Gap as first author on arXiv, showing that stale-memory over-trust failures in persistent-memory agents are gated by model capability.
- 2026Published The Structure of Quantization Damage in LLMs as first author on arXiv, finding damage is diffuse and global g128 granularity beats oracle local repair in models up to 8B.
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 indicators come from an OpenAlex record (identity: openalex) reporting h-index 0, i10-index 0, and only 2026 counts; recent arXiv preprints may not yet be fully indexed.
- No institution, country, or ORCID could be resolved; the profile is name-resolved from public bibliographic data, and Jundong Hu may be a common name.
- Output, impact, focus, and reproducibility radar axes are unscored because the underlying data is insufficient.
- All reviewed papers are 2026 arXiv preprints; the catalog contains no peer-reviewed venue record, citation data, or funding information.