LLM agent systems researcher at Salesforce working on long-horizon agent architectures and trace-based observability
2
papers here
1
reviewed by Evolve
1
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—
Not yet scored: this axis measures total publication output — work counts, venue types, and first/corresponding-author share — but the OpenAlex record is too sparse to compute reliably, with only six recent works and no aggregated output statistics.
Impact—
Not yet scored: this axis would measure citation impact via h-index, i10-index and citation counts over time. The snapshot lists h-index 0 and i10-index 0 for a very recent record, which is not yet a meaningful impact signal.
Focus100/100
Scored 100/100, the highest concentration on the radar. The OpenAlex field distribution lists a single field — 'AI agents (LLM systems)' — and the five assigned topics (topic modeling 3, NLP techniques 2, information retrieval 2, AI-based problem solving and planning 2, software system performance 2) all orbit agent infrastructure. Both 2026 preprints target long-horizon LLM agents, one on harness architecture and one on trace-based observability.
Collaboration28/100
Scored 28/100, reflecting a deliberately small footprint: OpenAlex records 3 distinct co-authors in the recent window, and Evolve names Erik Nijkamp, Anurag Koul and Bo Pang as recurring collaborators. Positions differ across the two catalog papers — first author on the live trace model, middle author on the long-horizon architecture, and not corresponding on either — suggesting focused small-team work rather than broad consortium building.
Activity57/100
Scored 57/100 on sustained recent activity. OpenAlex counts 6 works in the trailing three years, rising from 2 in 2025 to 4 in 2026, the strongest year in the snapshot. Both Evolve-catalog papers appeared as 2026 arXiv preprints within this uptick, and the works stay within the agent-systems topic cluster.
Reproducibility20/100
Scored 20/100, computed from 1 assessed paper with a mean readiness of 20. The long-horizon architecture paper is rated low readiness — a single 10-day autonomous RL reproduction with daily human checks and no baselines — while the trace-model paper reaches medium readiness with controlled 30/30 vs 8/30 comparisons but co-developed benchmarks. Independent replication evidence is absent.
Research summary
Egor Pakhomov is a US-based researcher at Salesforce whose profile is resolved from OpenAlex bibliographic data; no ORCID is linked in this snapshot. The record is early-stage: OpenAlex lists an h-index and i10-index of 0 and six works over the trailing three years, with 2 works in 2025 and 4 in 2026. Assigned topics concentrate on topic modeling, natural language processing techniques, information retrieval and search behavior, AI-based problem solving and planning, and software system performance and reliability, and the field distribution is dominated by AI agents (LLM systems). Evolve has reviewed two 2026 arXiv preprints. The first-authored 'Parsing the Stream' proposes a live trace model that serves a long-horizon agent and its observers from one context fold, with controlled comparisons reported at 30/30 vs 8/30 and a 14x token cost, while 'An Architecture for Long-Horizon Agents' introduces a levels/ticks/cascaded-intelligence harness demonstrated on a 10-day autonomous RL reproduction with daily human checks. Frequent co-authors include Erik Nijkamp, Anurag Koul and Bo Pang. Citation impact, output volume and reproducibility signals remain thin, so conclusions should be treated as provisional.
Collaboration pattern
The record shows a small, stable collaboration footprint: OpenAlex lists 3 distinct co-authors in the recent window, and Evolve's catalog names Erik Nijkamp, Anurag Koul and Bo Pang as frequent co-authors at Salesforce. Authorship positions vary — first author on the live trace model paper, middle position on the long-horizon architecture paper, neither corresponding — consistent with a contributor role inside a small LLM-agents group rather than an independent lab lead. Collaboration score: 28/100.
Research keywords
Highlights
10-day autonomous RL reproduction as an existence proof
One trace stream serves both agent and observer
Observability treated as a first-class agent problem
Timeline
- 2025Earliest year in the OpenAlex snapshot: 2 works indexed, opening the three-year output window that totals 6 works.
- 2026First-authored arXiv preprint 'Parsing the Stream: A Live Trace Model for Long-Horizon Agents and Their Observers', rated medium reproducibility readiness.
- 2026arXiv preprint 'An Architecture for Long-Horizon Agents: Levels, Ticks and Cascaded Intelligence' (middle-position author), rated low reproducibility readiness.
- 2026Strongest year in the snapshot: 4 works indexed, completing 6 works over the trailing three years.
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 are drawn from OpenAlex only; no ORCID is linked, so name resolution and work attribution may be incomplete.
- h-index and i10-index are 0 in this snapshot, so output and impact dimensions are unscored and any citation-based comparison would be premature.
- Both reviewed papers are 2026 arXiv preprints; peer-reviewed venue acceptance is not represented in this data.
- Reproducibility signals cover 1 paper (mean readiness 20); the rest of the record is unassessed.