Open full-duplex speech-to-speech systems and LLM post-training toward gold-medal competitive coding
2
papers here
2
reviewed by Evolve
0
first-author
0
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—
The output axis is not scored: the current snapshot provides no publication-count or productivity data for this researcher. Once cumulative paper counts and yearly output volumes are ingested, this axis would measure how much peer-reviewed and preprint work the researcher produces over time.
Impact—
The impact axis is not scored: no citation counts, h-index, or influence metrics are available in this snapshot. It would measure how widely the researcher's work is cited and adopted by other papers, models, and downstream systems.
Focus0/100
The focus axis scores 0, reflecting even dispersion rather than missing data: the raw field distribution splits exactly 0.5 speech processing and conversational AI versus 0.5 machine learning (LLM reasoning and code generation). The two reviewed 2026 papers sit one in each camp — an open full-duplex speech-to-speech model and a post-training pipeline for competition coding — so current cataloged output spans two distinct research communities in equal measure.
Collaboration83/100
Collaboration scores 83, backed by 62 distinct coauthors in the snapshot — a scale typical of large industrial research groups. The twelve most frequent collaborators skew toward the speech strand (Jagadeesh Balam, Travis Bartley, Edresson Casanova, Zhehuai Chen), and both reviewed papers carry long multi-author lists. Placement — middle author on VoiceChat and last author on the IOI post-training paper — fits senior contributions inside team-written work rather than solo output.
Activity—
The activity axis is not scored: the snapshot lacks year-by-year publication records beyond the two reviewed 2026 papers. It would measure recent publishing cadence and continuity of research engagement across consecutive years.
Reproducibility55/100
Reproducibility averages 55 across the 2 papers scored so far, both rated medium readiness. The VoiceChat model is released openly, which aids replication, but its design claims are untested and argument accuracy is weak (42.2%, Pass@1 33.0%); the IOI pipeline's headline 535.4/600 result rests on a single live unsupervised run. Open artifacts support reproduction, yet key results still await independent verification.
Research summary
Boris Ginsburg is a researcher working at the junction of spoken conversational AI and large-language-model post-training; this profile is name-resolved from public bibliographic records, with affiliation, country, and ORCID unconfirmed in the underlying data, though the reviewed papers are associated with NVIDIA's Nemotron work. The Evolve catalog holds two 2026 arXiv papers. The first, NemotronLabs VoiceChat, is an open full-duplex speech-to-speech model with native tool calling, praised for strong turn-taking and barge-in behavior but showing weak argument accuracy (42.2%, Pass@1 33.0%) and untested design claims. The second describes an SFT, GRPO-based reinforcement learning, and GenCorrect pipeline that reaches IOI gold-medal level; a live unsupervised IOI 2026 run scored 535.4/600, above the top human competitor, though it rests on a single run. A 50/50 raw field split between speech processing and conversational AI and machine learning (LLM reasoning and code generation) indicates a deliberate two-track agenda. With 62 distinct coauthors and a core of twelve frequent collaborators, the work emerges from large team settings; Ginsburg is a middle author on the speech system paper and last author on the coding paper. Reproducibility averages 55 across the two scored papers, both medium readiness. Citation, output-volume, and activity metrics are not yet available.
Collaboration pattern
Work is produced inside large teams: 62 distinct coauthors appear in the snapshot, with a core of twelve frequent collaborators largely from the speech strand (Jagadeesh Balam, Travis Bartley, Edresson Casanova, Zhehuai Chen). Authorship positions — middle on VoiceChat, last on the IOI post-training paper — suggest a senior, supervisory role in group-written industrial research rather than solo publication, bridging the speech and LLM-reasoning collaborator pools.
Research keywords
Highlights
First open full-duplex speech-to-speech model with native tool calling
Post-training an LLM to IOI gold-medal level
Timeline
- 2026Co-authored (middle position) NemotronLabs VoiceChat, an open full-duplex speech-to-speech model with native tool calling capabilities.
- 2026Last-authored the post-training paper whose SFT+RL+GenCorrect pipeline reached IOI gold, with a live unsupervised IOI 2026 run scoring 535.4/600.
- 2026Both catalog papers appeared in the same year, pairing open speech-system engineering with LLM reasoning and code-generation post-training.
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.
- Dimension scores come from a limited bibliometric snapshot; the output, impact, and activity axes lack underlying data and are not scored.
- Institution, country, and ORCID are absent from the profile data; identity is name-resolved from public bibliographic records, and the reviewed papers are associated with NVIDIA's Nemotron work.
- Only two reviewed papers are in the Evolve catalog, so this profile may not reflect the researcher's full publication record.
- The reproducibility score of 55 is computed from only 2 papers, both rated medium readiness.