JV / RESEARCH
Javal Vyas

Javal Vyas

PhD student
Imperial College London
Curriculum vitae

London, UK
Optimization & machine learning

GitHub ↗LinkedIn ↗
RESEARCH / ENGINEERINGIMPERIAL COLLEGE LONDON

Javal VyasOptimization &
machine learning.

I develop optimization algorithms, machine learning methods, and Python research software for decision-making under constraints. My work spans energy-system modelling, scheduling, and agentic control.

PhD · Chemical EngineeringSep 2024 — Dec 2027 (expected)Google Scholar ↗ Curriculum vitae ↗

Research interests Quantitative research / Development / AI

01 Selected work

MODELS → METHODS → SOFTWARE
01

Mathematical modelling · Energy

Unit Commitment Optimization

Optimization models and decomposition algorithms for meeting electricity demand at minimum cost under combinatorial commitment constraints.

4× max. COMPUTATIONAL SPEEDUP
0.18% median OBJECTIVE DEVIATION

SP3120 · Tight formulation
Separate benchmark statistics.
Method & results, Table 3 ↗

Energy SystemsOptimizationScheduling
Methods & contribution
  • Shrinking-horizon decomposition with relaxed future binary decisions, compared with monolithic EGRET formulations
  • Evaluated on four power-system cases with 20 instances per case
  • SP3120, Tight formulation: up to 4× speedup and 0.18% median objective deviation, reported as separate benchmark statistics
  • Results vary by case and formulation; CASE6468RTE has larger deviations and time-limited reference solves
Paper ↗
02

Open-source software · Mathematical optimization

PARETO: Surrogates for Water Optimization

Contributions to an open-source framework for strategic produced-water management, integrating desalination surrogate models into optimization workflows.

PythonPyomoSurrogate Models
Methods & contribution
  • Linked machine-learning surrogates to treatment planning and cost optimization
  • Co-authored the 2024 PARETO capabilities paper

My contribution. Contributed membrane-distillation and mechanical-vapor-compression surrogate integration, a training notebook, network visualization, and data-validation improvements.

Paper ↗ Code ↗
03

Industrial collaboration · AI evaluation

ADMITBench

A reference framework for assessing whether industrial LLM advisories satisfy explicit evidence, authority, procedure, and consequence checks.

LLM EvaluationConstraintsAdmissibility
Methods & contribution
  • Separates action eligibility from utility ranking using mandatory checks
  • Technical white paper and reference implementation for research evaluation

My contribution. Co-authored through the collaboration between Imperial College London and Refiant, Inc.

Paper ↗
04

AI systems · Control

Fault-Tolerant Control with LLM Agents

Agentic decision framework that turns fault signals into constraint-aware recovery plans, then checks candidate actions through simulation and deterministic validators.

LLM AgentsValidationControl
Methods & contribution
  • Multi-agent monitoring, planning, action synthesis, simulation, and reprompting
  • Evaluates action reliability under process constraints before execution

My contribution. Built most of the system, including the agent architecture, simulation, validators, and experiments. Collaborators developed the knowledge graph and SPARQL queries.

Paper ↗ Code ↗
05

Multimodal AI · Graphs

P&ID to Process Graphs

Multimodal language-model workflow for extracting equipment tags and reconstructing process topology from P&ID drawings.

Multimodal MLGraphsStructured Data
Methods & contribution
  • Separates visual extraction from topology reasoning
  • Targets scalable, semantically reliable P&ID digitization
Paper ↗
06

Open-source software · Optimization

rtn_scheduling

Python package for solving resource-task-network scheduling problems with Pyomo, including experiment and visualization utilities.

PythonPyomoOptimization
Methods & contribution
  • Resource-task-network inputs
  • Gantt, resource-level, and network visualizations
Code ↗

02 Publications

GOOGLE SCHOLAR ↗
  1. 2026

    ADMITBench: A Safety-Governed Reference Framework for Evaluating the Admissibility of Industrial LLM Advisories

    arXiv / technical white paper

    ↗
  2. 2026

    Automating Cause-Effect Specification with Knowledge Graphs and Large Language Models

    arXiv / preprint

    ↗
  3. 2026

    From Detection to Action: Using LLM Agents for Fault-Tolerant Control

    arXiv / preprint

    ↗
  4. 2026

    From P&ID Drawings to Process Graphs: A Multimodal Language Model Approach

    Systems and Control Transactions / journal article

    ↗
  5. 2025

    Optimization models and algorithms for the Unit Commitment problem

    Systems and Control Transactions / conference proceedings

    ↗
  6. 2024

    An Update on Project PARETO - New Capabilities in DOE's Produced Water Optimization Framework

    FOCAPD · Systems and Control Transactions / conference proceedings

    ↗
  7. 2024

    Integration of Plant Scheduling Feasibility with Supply Chain Networks Under Disruptions Using Machine Learning Surrogates

    ESCAPE / conference

    ↗

03 Research & experience

FULL CV ↗

Sep 2024 – Dec 2027 (expected)

Imperial College London

Research in optimization, machine learning, and autonomous industrial systems.

Ongoing

Imperial College London × Refiant, Inc.

Co-authored ADMITBench, a reference framework for evaluating industrial LLM advisories. Ongoing collaboration explores long-context models and frameworks that use prior knowledge to assess action admissibility.

ADMITBench white paper ↗

04 Methods & expertise

I work on decision-making under constraints: modelling complex systems, learning from data, and evaluating whether proposed actions are feasible.

Interested in quantitative research and development, with additional interests in AI/ML research and research engineering.

Research Engineering

Python / Git / Linux / Docker / CI/CD

Optimization

Optimization / Scheduling / MILP / MINLP / Decomposition / Experiment design

ML Systems

Machine learning / Uncertainty diagnostics / Validation loops / Surrogates / Model evaluation

Agents / Knowledge

Agentic workflows / Retrieval / Graph RAG / Knowledge graphs / Tool use / Semantic constraints

Systems Domain

Process control / Fault handling / Energy systems / Digital twins / P&ID digitization

05 Contact

LONDON, UK
javalvyas2000@gmail.com

Quantitative research & development · AI/ML research · Research engineering

GitHub ↗LinkedIn ↗Medium ↗Google Scholar ↗