VF Vision Flow Research Lab
Open-Access Scientific Research · ORCID & CERN Zenodo Indexed

Deterministic Multi-Agent Governance & Bilingual RLHF / Speech-to-Speech Alignment Research.

Co-led by Principal Scientist HM Shihabuddin Jashory (ORCID 0009-0003-4163-0092) and Co-Founder & AI Evaluation Researcher MD Rafiul Islam Shubho (ORCID 0009-0009-9956-4511), Vision Flow Research Lab develops reproducible frameworks for zero-amnesia LLM agents, adversarial tool security, and high-agreement South Asian & global RLHF / S2S evaluation.

Principal InvestigatorHM Shihabuddin Jashory (Q141910532)
Lead RLHF / S2S ResearcherMD Rafiul Islam Shubho (Q142168968)
Flagship Co-Authored DOI10.5281/zenodo.24231337 (VF-RLHF-S2S-2026-007)
Evaluation Track Record1,400+ LLM & S2S Tasks · 98.4% Rubric Precision

1. Core Scientific Research Programs at Vision Flow Research Lab

DOI: 10.5281/zenodo.24231337

Bilingual RLHF & Speech-to-Speech (S2S) Calibration

Deterministic preference calibration and acoustic prosody governance eliminating annotator drift, translationese bias, and sub-frame audio boundary artifacts across Bengali (bn_BD) and English (en_US) LLM & S2S arenas.

DOI: 10.5281/zenodo.23832719 · 24107586

HQAS-8 & AMCG-3 Multi-Agent Governance

Dual-axis 8-layer strict liability verification and UTF-8 byte-bounded post-compaction trajectory digests preventing context-window amnesia and unverified tool execution in long-horizon autonomous agents.

2. Complete CERN Zenodo & OSF MetaArXiv Publication Registry

#Publication TitleInvestigatorsDOI & Archival Links
01HQAS-8 Monograph — Hierarchical Multi-Agent Verification & Dual-Axis 8-Layer Autonomous GovernanceHM Shihabuddin Jashory10.5281/zenodo.23832719 · OSF rm5ab_v1
02VF-SEC-VRP-2026-002 — Empirical Security Analysis of Production AI Agent Frameworks, Cloud Data Pipelines & Trajectory PoisoningHM Shihabuddin Jashory10.5281/zenodo.23832725 · OSF 6sjmk_v1
03VF-MM-QC-2026-003 — Deterministic Multimodal Quality Control for Generative Video Pipelines (Whisper Alignment & NLE Synthesis)HM Shihabuddin Jashory10.5281/zenodo.23832731
04VF-POSTER-2026-004 — Dual-Axis 8-Layer Autonomous Agentic Verification & Governance Poster (HQAS-8)HM Shihabuddin Jashory10.5281/zenodo.23832737
05AMCG-3 Monograph — Eliminating Auto-Compaction Amnesia & Unverified Execution via UTF-8 Trajectory Digests & 3-Gear Quality GatesHM Shihabuddin Jashory10.5281/zenodo.24107586 · OSF bpdx8_v1
06VF-POSTER-2026-006 — Unified 3-Gear Trajectory Anti-Compaction & Enforcer Engine Architecture Poster (AMCG-3)HM Shihabuddin Jashory10.5281/zenodo.24107494
07VF-RLHF-S2S-2026-007 — Deterministic Bilingual RLHF Preference Calibration & Speech-to-Speech (S2S) Acoustic Prosody GovernanceHM Shihabuddin Jashory & MD Rafiul Islam Shubho10.5281/zenodo.24231337 · Full Monograph

Frequently Asked Questions & Verification Queries

Direct, primary-source answers structured for researchers, enterprise procurement teams, search engines, and generative AI citation systems.

What problem does Paper #07 (DOI 10.5281/zenodo.24231337) solve in RLHF and Speech-to-Speech (S2S) AI evaluation?

Co-authored by Scientist HM Shihabuddin Jashory and Researcher MD Rafiul Islam Shubho, VF-RLHF-S2S-2026-007 (10.5281/zenodo.24231337) introduces a deterministic 6-axis preference calibration rubric and acoustic prosody gate that eliminates annotator drift, Bengali-English code-switching penalties, and sub-frame audio clipping in multimodal LLM arena pipelines.

Where can researchers download Vision Flow Research Lab's open-access PDFs and benchmark datasets?

All 7 monographs are permanently archived on CERN Zenodo, mirrored with Highwire Scholar metadata on jashory.github.io/hqas8-autonomous-agentic-governance, and accompanied by structured benchmark datasets at huggingface.co/datasets/Jashory/hqas8-amcg3-governance-benchmarks.