ML researcher (Liverpool John Moores University) in synthetic data generation and efficient video-language models
2
papers here
2
reviewed by Evolve
2
first-author
0
corresponding
—
works · OpenAlex
—
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-15) 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 metrics are not yet available for this profile. This axis would measure cumulative publication volume, venue distribution and first-author share over time, none of which can be computed from the current Evolve catalog of two 2026 preprints.
Impact—
No citation or impact data is available yet. This axis would measure citation counts, h-index-style indicators and uptake of the two 2026 preprints, which are too recent and unindexed for citations in the current data.
Focus0/100
The focus score sits at 0 because recorded output is split exactly evenly between two distinct subfields — machine learning for synthetic time-series generation in manufacturing and machine learning for multimodal video-language models (0.5 and 0.5 in the field distribution). Both 2026 papers are first-author, suggesting deliberate parallel tracks rather than drift, but the agenda has not yet consolidated around a single line of work.
Collaboration41/100
Collaboration scores 41 on a 0–100 scale, based on 7 distinct coauthors across the two catalogued papers. The team is compact and recurring — Jeshwanth Challagundla, Prateek Karnal, Gagan Ganapathy, Vineet Shah, Ridam Arora, Siddharth Raina and Jasmin Jarsania — and Singh takes first authorship on both 2026 preprints without corresponding-author credit, indicating a hands-on lead-experimentalist role within a small, stable group.
Activity—
Activity data is insufficient in the current snapshot. This axis would measure publishing cadence, preprint-to-publication turnaround and ongoing project throughput over time; with only two 2026-dated preprints catalogued, any cadence estimate would be premature.
Reproducibility38/100
Reproducibility scores 38, the mean readiness rating across 2 assessed papers. Evolve rates the video-language distillation paper medium — the result of a 2B model nearing its 4B teacher from 900 synthetic CoT samples is promising but single-run, with no variance and an unquantified CoT-order ablation — and the manufacturing time-series paper low, due to private data, weak baselines and no significance tests. Releasing code, data and statistical detail would lift this axis.
Research summary
Mantek Singh is a machine learning researcher at Liverpool John Moores University whose early Evolve-indexed catalog centres on one recurring idea: using generative models to synthesise training data where real data is scarce or expensive. Across two first-author arXiv preprints dated 2026, he pushes that idea in two directions. In manufacturing, he and co-authors use large language models with retrieval-augmented generation (GPT-3.5 Turbo) to generate synthetic time series, claiming better results than ARIMA and LSTM baselines; Evolve rates the study plausible but thin, citing weak baselines, no significance tests and private data. In multimodal learning, he leads reasoning-distillation work showing that 900 synthetic chain-of-thought samples with difficulty-aware, uncertainty-based selection can bring a 2B video-language model near its 4B teacher; the Evolve review flags single-run results with no variance and an unquantified CoT-order ablation. His reproducibility score of 38 (mean across 2 assessed papers, readiness rated medium and low) and an even 50/50 split between his two subfields point to an exploratory, early-stage agenda. Collaboration is compact: 7 distinct coauthors, including Jeshwanth Challagundla, Prateek Karnal and Gagan Ganapathy, with first-author credit on both papers but no corresponding-author role. Output, impact and activity metrics are unavailable, so this profile should be read as an early-stage snapshot rather than a mature record.
Collaboration pattern
The catalogued work is a compact-team effort: 7 distinct coauthors, all recurring — Jeshwanth Challagundla, Prateek Karnal, Gagan Ganapathy, Vineet Shah, Ridam Arora, Siddharth Raina and Jasmin Jarsania. Singh is first author on both 2026 preprints but holds no corresponding-author credit, consistent with a lead-experimentalist role inside a small, stable group spanning synthetic manufacturing data and multimodal video-language modeling. A collaboration score of 41 reflects this early, low-breadth network.
Research keywords
Highlights
Synthetic CoT distillation brings a 2B video-language model near its 4B teacher
LLM+RAG synthetic time series for manufacturing, positioned against ARIMA and LSTM
Timeline
- 2026Publishes a first-author arXiv preprint, 'LLMs as Master Forgers', applying LLM+RAG (GPT-3.5 Turbo) to synthetic time-series generation for manufacturing.
- 2026Publishes a first-author arXiv preprint on efficient reasoning distillation, showing synthetic CoT and difficulty-aware fine-tuning bringing a 2B video-language model near its 4B teacher.
- 2026Both papers enter the Evolve catalog, establishing an even 50/50 split between synthetic-data manufacturing and multimodal video-language research, with 7 distinct coauthors.
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 dimensions are computed from OpenAlex-derived data and the Evolve-reviewed catalog only; citation and venue metrics are not yet available for these 2026 preprints.
- The output, impact and activity axes are missing from the current snapshot; no publication counts, citation counts or publishing-cadence figures are quoted.
- Only two first-author arXiv preprints (both 2026) are assessed; this profile is an early-stage snapshot, not a complete publication record.
- Identity is resolved via ORCID 0000-0003-3043-1461 from public bibliographic data; 'Mantek Singh' is not a unique name, so attribution rests on this identifier.