Works on recursive self-improvement of AI research agents and agentic search for LLM pre-training data selection
2
papers here
2
reviewed by Evolve
1
first-author
0
corresponding
—
works · OpenAlex
—
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-02) 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—
Not yet scored: publication volume, venue mix, and yearly output counts are absent from the underlying data, so productivity cannot be measured beyond the two 2026 arXiv papers in the Evolve catalog.
Impact—
Not yet scored: citation counts, h-index, and downstream uptake are missing from the snapshot; this axis would measure the citation-based influence and adoption of the work, and no estimate is made here.
Focus0/100
The focus score is 0, the lowest reading, and tracks the raw field distribution directly: the two cataloged papers split exactly 50/50 between autonomous research agents (the AIDE² work) and LLM pre-training data curation (AutoData). With a two-paper sample, this even split reads as a deliberate two-front agenda rather than scattered output, but it is too small to show whether one line will consolidate.
Collaboration41/100
The collaboration score is 41, computed from a compact pool of 7 distinct coauthors across the two cataloged papers; frequent names include Bingchen Zhao, Yuxiang Wu, Zhengyao Jiang, Dixing Xu, and Yan Meng. The pattern points to small-team, engineering-heavy projects rather than large consortia: first-author lead on AIDE², a middle position on AutoData, and corresponding-author credit on neither.
Activity—
Not yet scored: no publication cadence or year-over-year coverage exists beyond both papers sharing 2026, so sustained-output trends cannot be measured; this axis would track research activity over time.
Reproducibility38/100
The reproducibility score is 38, the mean over the two reviewed papers. AIDE² rates low readiness — Evolve flags few runs and noisy grade comparisons behind the 0.703→0.778 improvement — while AutoData rates medium, its val-bpb gains holding across scales even though CORE gains lack statistical support. Statistical rigor and artifact release are the clear growth area.
Research summary
Dhruv Srikanth is an early-career researcher whose cataloged work sits at the intersection of autonomous AI research agents and LLM pre-training data curation. The Evolve catalog holds two 2026 arXiv papers, splitting the field distribution evenly at 50/50 between the two areas. The first-authored paper, Recursive self-improvement of AI research agents, presents AIDE²: an eight-day self-improvement loop that lifts agent grade from 0.703 to 0.778 with held-out transfer. Evolve finds the result compelling but resting on few runs and noisy grade comparisons, and rates reproducibility readiness low. The second paper, AutoData: Agentic Search for Pre-training Data Selection, lists Srikanth in a middle author position; it shows LLM-agent search over executable data-selection recipes beating curated-pipeline baselines on validation bits-per-byte across scales, with medium reproducibility readiness, though the reported CORE gains lack statistical support. Bibliometrically the footprint is compact: 7 distinct coauthors, a collaboration score of 41, a focus score of 0 reflecting the even two-field split, and a reproducibility mean of 38 across the two reviewed papers. Output, impact, and activity axes currently lack sufficient data to score. No institution, country, or ORCID is recorded; the profile is name-resolved from public bibliographic data.
Collaboration pattern
The cataloged footprint follows a small-team pattern: 7 distinct coauthors across two 2026 arXiv papers, with a collaboration score of 41. Bingchen Zhao, Yuxiang Wu, Zhengyao Jiang, Dixing Xu, and Yan Meng recur among frequent coauthors. Srikanth takes first authorship on the AIDE² agent work and a middle position on AutoData, with no corresponding-author credit in the catalog — consistent with contributing core engineering inside a compact, possibly shared research group.
Research keywords
Highlights
AIDE²: an 8-day recursive self-improvement loop for AI research agents
AutoData: agentic search over executable data-selection recipes
Timeline
- 2026Publishes AutoData (arXiv, middle-author position), applying agentic LLM search to pre-training data selection with validation-bpb gains across scales.
- 2026First-authors Recursive self-improvement of AI research agents (arXiv), presenting AIDE²'s 8-day loop that lifts agent grade 0.703→0.778 with held-out transfer.
- 2026Both cataloged papers are reviewed in the same year, splitting evenly across autonomous research agents and LLM pre-training data curation, with reproducibility readiness of low and medium respectively.
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 scores are computed from a small public snapshot: two reviewed papers and 7 distinct coauthors, with no citation or publication-history data.
- Output, impact, and activity radar axes are unscored because the underlying volume, citation, and cadence data is not available.
- No institution, country, or ORCID is recorded; the profile is name-resolved from public bibliographic data, so common-name ambiguity cannot be fully excluded.
- Both cataloged papers are 2026 arXiv preprints; peer-reviewed venue status and final versions are unknown.