Chandra Bhushan

Banaras Hindu University · India

Unclaimed · AI-structured profileORCID OpenAlex

LLM-agent researcher at Banaras Hindu University: evaluation benchmarks and post-training for Indian retail banking

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 OpenAlex snapshot supplies only an h-index of 9 and i10-index of 9, without complete per-venue and per-year publication counts, so Evolve cannot compute a reliable output score for this researcher yet.

Impact—

No score yet. This axis would measure normalized citations, citation percentiles against field and year, and the influence of the reviewed papers; the snapshot carries only h-index 9 and i10-index 9, which is insufficient to compute it.

Focus0/100

Scores 0 on topical concentration. The OpenAlex topic profile is split across at least five clusters — Agricultural Science and Fertilization (21 works), Helminth infection and control (15), Agronomic Practices and Intercropping Systems (10), Weed Control and Herbicide Applications (8), and Vector-borne infectious diseases (7) — while the reviewed LLM-agent papers sit outside all of them, so recorded research energy is highly dispersed across unrelated domains.

Collaboration60/100

Scores 60. Over the last three years this researcher coauthored 21 works with 19 distinct coauthors, and the frequent-coauthor list is anchored by the NPCI AI Research Team plus a stable cluster of individuals — Keyur Doshi, Prashant Devadiga, Suvradip Paul, Yatharth Dedhia — consistent with large team-authored arXiv papers in which this researcher holds a middle, non-corresponding position.

Activity90/100

Scores 90, the strongest axis. The countsByYear series shows 4 works in 2024, 9 in 2025, and 8 already in 2026, with recent3yWorks totaling 21 — a marked acceleration over the 1–3 works per year typical of most earlier years in the record, with the earlier 2017 peak of 9 matched again in 2025.

Reproducibility60/100

Scores 60, averaged over two Evolve-reviewed papers (reproMean 60). IndicBankBench is rated high reproducibility readiness — a staged benchmark reporting both strict and at-least-once pass rates (43.7–58.2% vs 60–74%) — while FiMI Banking is rated low, using a self-authored evaluation and releasing no model or environment, pulling the mean to the midpoint.

Research summary

Chandra Bhushan (Banaras Hindu University, India; ORCID 0009-0003-2351-3300) is a researcher whose Evolve-reviewed work concentrates on large language model agents for Indian retail banking. Two 2026 arXiv papers define the current line. IndicBankBench is a staged safety and reliability benchmark showing that at-least-once success (60–74%) overstates agent reliability relative to strict pass rates (43.7–58.2%); Evolve rates its reproducibility readiness high while noting a thin judge audit. FiMI Banking trains a 4.5B-parameter banking agent with DPO and GRPO in a self-built synthetic environment, raising reward from .610 to .697 and refusal rate from 52% to 80%, though it relies on self-authored evaluation and no release. The bibliometric snapshot is active — 21 works in the last three years with 19 distinct coauthors, h-index 9, i10-index 9 — but OpenAlex topic clusters (agricultural science, helminth infection control, weed control) diverge from the reviewed AI papers, so this profile is name-resolved from public bibliographic data. The record spans 1987–2026 with a recent surge of 9 outputs in 2025 and 8 in 2026; output and impact dimensions remain unscored pending fuller data.

Collaboration pattern

Collaboration is team-based: 19 distinct coauthors across 21 works in the last three years. The frequent-coauthor list centers on the NPCI AI Research Team and recurring individuals — Keyur Doshi, Prashant Devadiga, Suvradip Paul, Yatharth Dedhia — consistent with industry-linked, large-author-group arXiv papers. On both reviewed papers this researcher appears in a middle position without corresponding authorship, suggesting a contributor role inside a stable banking-AI team rather than sole leadership.

Research keywords

LLM agentsbenchmark designagentic tool useagent reliability evaluationpass@k evaluationLLM-as-judgepost-trainingGRPOdirect preference optimization (DPO)reinforcement learning with verifiable rewardssynthetic data generationIndic retail banking NLP

Highlights

Strict pass rates undercut at-least-once claims in banking agents
Evolve's review of IndicBankBench found that at-least-once success rates of 60–74% overstate banking-agent reliability, with strict pass rates falling to 43.7–58.2%. The staged benchmark is rigorous, though the LLM-as-judge audit is thin; reproducibility readiness is rated high.
DPO+GRPO post-training lifts a 4.5B banking agent
FiMI Banking trains a sovereign 4.5B-parameter banking agent in a self-built synthetic environment: reward rises from .610 to .697 and refusal rate from 52% to 80%. Evolve calls the method solid but flags the self-authored evaluation and the absence of any release, rating reproducibility readiness low.
A paired evaluation-plus-training loop for Indic banking agents
The two 2026 papers complement each other: IndicBankBench supplies an external staged benchmark for Indian retail banking assistants, while FiMI Banking builds and trains the agent itself. Together they outline a full build-evaluate-train cycle, though the benchmark's judge audit and the model's release remain open gaps.

Timeline

1987Earliest output recorded in the OpenAlex snapshot (1 work); this early record sits far from the current AI line and forms part of the name-resolution uncertainty.
2017Busiest earlier year in the record with 9 outputs, a peak not matched until 2025.
20233 outputs recorded, consistent with the steady 1–3 works per year typical of most years in the preceding stretch.
2025Output accelerates to 9 works in a single year, matching the 2017 peak.
20268 outputs so far; both Evolve-reviewed banking-agent papers appear — IndicBankBench (evaluation) and FiMI Banking (post-training).

Papers in the Evolve catalog

Frequent co-authors

Keyur Doshi×2 Prashant Devadiga×2 Suvradip Paul×2 Yatharth Dedhia×2 NPCI AI Research Team×1 Aman Kumar×1 Asit Desai×1 Harsh Sharma×1 Harshit Bhushan×1 Hrithik Kadam×1 Kolisetty Sai Kapardheeswar×1 Krishanu Adhikary×1

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 come from an OpenAlex snapshot only; no Scopus, Web of Science, or Google Scholar data.
Output and impact dimensions are unscored (insufficient data) — the snapshot lacks the complete corpus needed to compute them.
Topic clusters (agriculture, helminth infection, weed control) diverge sharply from the reviewed LLM-agent papers; "Chandra Bhushan" is a common name, so this profile is name-resolved from public bibliographic data and may mix records of distinct individuals.
Both reviewed papers are 2026 arXiv preprints with middle, non-corresponding authorship; peer-review status and citation uptake are not yet reflected in the metrics.