Mahesh Godavarti

—

Unclaimed · AI-structured profileOpenAlex

Emerging machine learning researcher of group-theoretic and content-based positional addressing for transformers

papers here

reviewed by Evolve

first-author

corresponding

—

works · OpenAlex

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—

Not yet scored: the snapshot records only 10 recent works alongside an h-index and i10-index of 0, which is too thin to characterize research output. This axis would measure total publications and preprints plus yearly production volume.

Impact—

Not yet scored: citation signals are absent — the h-index and i10-index both read 0 — so influence cannot be assessed yet. This axis would measure citation counts, citing works, and uptake of the proposed addressing methods.

Focus0/100

The recorded field distribution is a clean 50/50 split between machine learning and machine learning for NLP (0.5 each), and the topic profile concentrates on Topic Modeling (8 assignments), NLP Techniques (5), and neural-network and multimodal applications (3+3+2). The substantive agenda is tightly focused on positional encoding and attention addressing, even though the composite focus score in this snapshot reads 0, likely reflecting the short two-year record.

Collaboration0/100

The snapshot records 0 distinct co-authors and lists no frequent collaborators, and both cataloged 2026 papers carry Godavarti as first and corresponding author — hence the collaboration score of 0. The visible pattern is an essentially solo research mode so far; a co-authorship network would only emerge as more works enter the index.

Activity70/100

Recent output is steady and accelerating: 10 works fall within the last three-year window, split as 4 outputs in 2025 and 6 in 2026, which underpins the activity score of 70. Both cataloged arXiv papers appeared in 2026, indicating that the current research rhythm is concentrated in the most recent year.

Reproducibility55/100

Evolve has assessed reproducibility for 2 papers with a mean score of 55, and both are rated medium readiness. The LIS paper relies on synthetic single-seed diagnostics, while the long-context addressing paper uses single-run, tiny-scale experiments; reviewers note the honest framing, but practical value remains unproven until results replicate at realistic scale.

Research summary

Mahesh Godavarti is a machine learning researcher whose profile is name-resolved from public bibliographic data (OpenAlex); no institutional affiliation is recorded in the current snapshot. The visible record is recent and compact: 10 works in the last three-year window, distributed as 4 outputs in 2025 and 6 in 2026, with the OpenAlex topic profile led by Topic Modeling (8 assignments), Natural Language Processing Techniques (5), and smaller clusters in neural-network and multimodal applications. The field distribution splits evenly between machine learning (0.5) and machine learning for NLP (0.5). Two 2026 arXiv papers in the Evolve catalog list Godavarti as both first and corresponding author, and they attack one underlying question: how transformers should address positions and units. The first introduces Labeled Incidence Structures, group-theoretic addresses that natively unify RoPE-style attention across text, knowledge graphs, and hypergraphs; the second replaces global rotary encodings with content-derived unit addresses for long-context modeling and length extrapolation. Reviewers judge the theory elegant and the framing honest but note that the evidence is synthetic, single-seed or single-run, and tiny in scale; reproducibility readiness is rated medium, with a mean of 55 across the 2 assessed papers. Citation-based signals (h-index 0, i10-index 0) are not yet informative for impact.

Collaboration pattern

In the current snapshot the collaboration footprint is empty: 0 distinct co-authors are recorded, no frequent co-authors are listed, and both cataloged 2026 papers carry Godavarti as first and corresponding author. The visible pattern is therefore solo-authored, theory-driven work on transformer addressing; with no institutional affiliation available, the team setting cannot be contextualized. A collaboration network has yet to form in the indexed record.

Research keywords

positional encodingrotary position embedding (RoPE)transformer attentioncontent-based addressinglong contextlength extrapolationknowledge graphshypergraphsstructured datatopic modelingmultimodal machine learningneural networks

Highlights

Group-theoretic addresses unify transformer attention across text, knowledge graphs, and hypergraphs
The 2026 LIS paper introduces Labeled Incidence Structures as group-theoretic addresses enabling native transformer modeling of text, knowledge graphs, and hypergraphs. Evolve's verdict: elegant theory, but the evidence is synthetic single-seed diagnostics; reproducibility readiness is medium.
Content-derived unit addresses replace global RoPE for long-context transformers
The companion 2026 paper replaces global rotary position embeddings with content-based unit addressing for long-context attention and length extrapolation. Reviewers find the theory clean and the framing honest, but single-run tiny-scale diagnostics leave practical value unproven; readiness is medium.
A consistent program: rethinking positional addressing in transformers
Both cataloged papers — each first-author and corresponding-author work from 2026 — attack the same question from different angles: replacing or generalizing RoPE. Both carry medium reproducibility readiness (mean 55 across 2 papers), and the evidence base is currently solo-authored, small-scale experimentation.

Timeline

2025Earliest year with recorded output in the snapshot: 4 works, according to OpenAlex counts-by-year.
2026Most productive recorded year: 6 outputs, including two arXiv preprints on transformer positional addressing.
2026Posts two first-author, corresponding-author arXiv papers — the Labeled Incidence Structures framework and content-based addressing for long context; both receive medium reproducibility readiness with a mean of 55.

Papers in the Evolve catalog

2026-09-18first authorcorresponding
Labeled Incidence Structures for Native Transformer Modeling of Text, Knowledge Graphs, and Hypergraphs
2026-09-07first authorcorresponding
Content-Based Addressing for Long Context

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 counts, topics, and field distribution come from OpenAlex only; the identity is name-resolved from public bibliographic data and may merge or split namesakes.
No institutional affiliation, country, or ORCID is available, so career context and team setting cannot be verified.
The visible record spans only 2025–2026 (10 works); h-index and i10-index read 0, so the citation-based output and impact axes are unscored.
Reproducibility rests on Evolve reviews of only 2 papers, both based on synthetic or single-run small-scale experiments.