Nieb Hasan Neom
Contact
WORK & RESEARCH

Systems, with the seams left visible.

Completed case studies and active research are separated by status. Each entry names the contribution, evidence, boundary, and next unknown.

Accepted research case studyCASE / GUARDED-ORCHESTRATION

Guarded Scientific Model Orchestration

A state-aware orchestration layer that extends LLM function calling with explicit workflow semantics so domain scientists can invoke trained prediction models through constrained natural language.

  • Accumulates required inputs across multiple turns and validates arguments against typed schemas.
  • Blocks model execution when workflow state is incomplete instead of guessing missing values.
  • Supports correction, what-if exploration, and reset steering while keeping execution boundaries visible.
LLM function callingSchema constraintsWorkflow stateScientific ML
Inspect case study
Completed backend case studyCASE / KONA-TOKEN-TRADE

Kona Token Trade

Backend service work for a blockchain-based token-trading platform, centered on reliable service communication, latency-conscious caching, and stronger controls for sensitive operations.

  • Contributed to trade, user, wallet, and fee-management services without claiming ownership of the full platform.
  • Introduced message queues to improve reliability between services and caching to reduce response latency.
  • Implemented two-factor authentication for sensitive trading operations; proprietary details and internal metrics remain excluded.
JavaSpring BootMicroservicesMessage queuesCachingTwo-factor authentication
Inspect case study
Completed research case studyCASE / BANGLA-SIGN-LANGUAGE

Bangla Sign Language Recognition

Undergraduate thesis research that contributed to a published hybrid recognition system combining image preprocessing, transfer learning, and a random-forest classifier.

  • Designed and evaluated the background-elimination preprocessing method.
  • Helped integrate preprocessing with the transfer-learning and random-forest recognition pipeline.
  • The multi-author system reported 91.67% character accuracy and 97.33% digit accuracy on open Bangla Sign Language datasets.
Image preprocessingTransfer learningRandom forestOpen datasetsExperimental evaluation
Inspect case study
Research in progressPrototype + evaluation

Agentic Data Transfer Optimizer

A bounded agentic system that adapts network-transfer concurrency from live TCP telemetry while making every recommendation, constraint, and state transition inspectable.

01 / Prototype implemented

Data-transfer runtime

A sender and receiver architecture uses multiprocessing workers and zero-copy file transfer over TCP.

02 / Evaluation in progress

Bounded agentic control

Diagnostic and Decision agents reason over live telemetry within hard concurrency limits and oscillation controls.

No completion or effectiveness claim until reproducible evidence is approved.Inspect research scope
PUBLICATIONS

Published, accepted, and preprint work—with status visible.

Google Scholar
2026

Guarded, State-Aware Orchestration for Schema-Constrained Scientific Model Execution

accepted21st Workshop on Workflows in Support of Large-Scale Science (WORKS26 at SC26)

2026

Self-Tuning Segmented Retrieval of Large-Scale Genomic Data

accepted22nd IEEE International Conference on e-Science (eScience 2026) · Short paper

2023

A hybrid approach for Bangla sign language recognition using deep transfer learning model with random forest classifier

publishedExpert Systems with Applications 213, article 118914

2019

A Review on Fall Detection Systems Using Data from Smartphone Sensors

publishedIngénierie des Systèmes d'Information 24(6), 569-576

2024

CredSec: A Blockchain-based Secure Credential Management System for University Adoption

preprintarXiv:2406.05151 · Preprint