Non-LLM (Non–Large Language Model) artificial intelligence systems

are not designed for natural language understanding or generation like

LLMs. Instead, they are typically built for narrow, domain-specific tasks
using rule-based, statistical, optimization-focused, or traditional
machine learning algorithms.
Non-LLM systems are highly effective in engineering tasks that require
structured data processing, numerical precision, deterministic control,
optimization, and prediction.
While LLMs are designed for human-like language generation and
contextual reasoning, Non-LLM systems serve as the core engines for
measurement, control, optimization, and forecasting in engineering.
While LLMs are trained with trillions of parameters and are capable of
human-like text generation,
Non-LLM systems:
• Require much less data
• Are faster and lighter to run
• Offer higher interpretability (e.g., decision trees)
• Focus on specific tasks rather than general-purpose reasoning
Non-LLM Systems
• These are usually traditional ML algorithms (decision trees, logistic
regression, support vector machines) or even rule-based expert
systems.
• Used when data is small and computational power is limited.
• They do not use deep neural networks → not Deep Learning.
• Example: A simple decision tree classifier for detecting defective
products in a factory line.
Non-LLM ≈ ML (mostly classical, shallow learning)
Category Description Example Algorithms / Techniques

Rule-Based Systems Encode human expertise using if-then rules Expert Systems, Production Rules

Support Vector Machines (SVM), Decision Tree,

Classical Machine Learning Learn from numerical data, not language
Random Forest, k-Nearest Neighbors

Probabilistic Models Make inferences based on statistical relationships Bayesian Network, Hidden Markov Model

Genetic Algorithm, Simulated Annealing, Linear

Optimization and Planning Algorithms Search for the best solution, not generative
Programming

Control and Signal Processing Systems Used in embedded systems for control/decision Kalman Filter, PID Controllers

Use domain-specific knowledge bases for

Knowledge-Based Systems Prolog-based expert systems
inference

Handle structured data like images or signals, not Convolutional Neural Network (CNN), Recurrent
Basic Deep Learning Models
text Neural Network (RNN)
Electrical and Electronics Engineering
• Fault diagnosis systems → Classify circuit failures using Decision Trees or
SVM
• Signal filtering and noise reduction → Use Kalman Filters
• Circuit design optimization → Parameter tuning via Genetic Algorithms

Mechanical Engineering
• Machining parameter optimization → Linear Programming
• Vibration analysis and fault prediction → Random Forest for prediction
• Robot kinematics control → PID controllers for precision control
Industrial Engineering
• Production line scheduling → Simulated Annealing, Tabu Search
• Demand forecasting → Time series models (ARIMA)
• Inventory optimization → Decision Trees, Linear Programming

Computer and Software Engineering

• Data mining and classification → SVM, Decision Trees
• Spam or malware detection → Naive Bayes
• Software defect prediction → Random Forest models
Civil Engineering
• Material strength prediction → Regression models
• Project time-cost optimization → Genetic Algorithms
• Structural health monitoring → Hidden Markov Models for anomaly
detection

Architecture
• Energy efficiency optimization → Multi-objective linear programming
• Thermal-flow simulations → Physics-based modeling + optimization
• Design evaluation → Rule-based systems for evaluating design variants
Industrial Design and Visual Communication Design
• User preference analysis → Clustering algorithms (k-Means)
• Color/shape trend analysis → PCA-based statistical analysis
• Product ergonomics scoring → Decision Trees for multi-criteria
evaluation

Interior Architecture and Environmental Design

• Lighting/acoustic simulations → Physics-based modeling
• Spatial layout optimization → Genetic Algorithms
• Cost estimation systems → Regression-based models
Urban and Regional Planning
• Traffic flow modeling → Agent-based simulations
• Land-use prediction → Markov Chains
• Environmental impact assessment → Multivariate statistical analysis
Advantages Limitations

Trainable with small datasets Narrow-task focused, not flexible

High interpretability Cannot understand complex context or language

Low hardware requirements Poor generalization to new tasks

Fast and real-time capable Hard to integrate large knowledge bases

These tools do not generate natural language, but instead perform
data processing, statistical modeling, optimization, classification,
control, and decision support tasks.
They are domain-specific and task-focused.
Rule-Based Tools
• CLIPS – Widely used for building expert systems
• Drools – Java-based business rules engine
• Prolog – Logic programming language for knowledge-based systems
Usage: Expert systems, fault diagnosis, decision support
Classical Machine Learning Tools
• Scikit-learn – Python-based classical ML library
• Weka – GUI-based data mining tool
• RapidMiner – Drag-and-drop ML platform
• Orange – Visual data mining and analytics
• KNIME – Workflow-based analytics platform
Usage: Classification, regression, clustering, anomaly detection
Statistical Analysis Tools
• R – Open-source statistical programming language
• SPSS – Widely used in social sciences for statistical analysis
• Minitab – Popular for quality engineering analysis
• Stata – Statistical and econometric modeling software
Usage: Time series analysis, correlation, variance, hypothesis testing
Optimization and Planning Tools
• Gurobi – Linear and nonlinear optimization solver
• CPLEX – Mixed-integer and scheduling optimization solver
• MATLAB Optimization Toolbox – Mathematical optimization
modules
• Simulated Annealing Algorithm – Metaheuristic algorithm (available
in many languages)
Usage: Production planning, route optimization, resource allocation
Control and Signal Processing Tools
• Simulink – System modeling and simulation
• LabVIEW – Hardware-based measurement and control applications
• NI Multisim – Electronic circuit simulation
• Kalman Filter – Signal noise reduction (available as libraries in
various languages)
Usage: Embedded system control, robotics, electronic circuit analysis
Basic (Non-LLM) Deep Learning Libraries
• TensorFlow
• Keras
• PyTorch
Note: These libraries can also be used to train LLMs, but when used
for CNNs, RNNs, or other structured-data models (not language
generation), they are considered Non-LLM tools.
Usage: Image/audio classification, time series forecasting, object
detection
Non-LLM tools:
• Do not generate text → work with numerical data
• Are domain-specific → focused on specialized engineering tasks
• Are explainable → results are mathematically traceable
• Complement LLMs → LLMs generate reports while Non-LLM tools
perform analytics/calculations
Aspect Non-LLM Tools LLM Tools

Main Purpose Task-specific analytical and predictive tasks Natural language understanding and generation

Data Type Structured / numerical / signal data Unstructured text and multimodal data

Training Data Requirement Small to medium datasets Massive datasets (billions/trillions of tokens)

Interpretability High – decisions are transparent and traceable Low – “black box” reasoning, hard to interpret

Generalization Ability Narrow, domain-specific Broad, cross-domain

Examples Scikit-learn, Weka, R, Simulink, Gurobi ChatGPT, Claude, LLaMA, Gemini

Predictions, classifications, control signals, Human-like text, code, documents, multimodal

Typical Output
optimized solutions outputs

Hardware Requirement Low (can run on desktop/embedded systems) High (requires GPU clusters)

Latency Very fast (real-time capable) Slower (depends on model size and inference)

Optimization, control, simulation, statistical Report generation, documentation, knowledge

Use Cases in Engineering
modeling retrieval, code generation

Built-in transparency (decision trees, regression Requires additional explainability frameworks

Explainability Tools
equations) (e.g., SHAP, LIME)
Prompt = User input / command.
In LLM-based systems (like GPT, Claude, LLaMA), prompts are free-
form natural language texts, and the model interprets them
statistically.
In Non-LLM tools, however, prompts are usually:
• selected from predefined command sets (menus, parameters,
functions), or
• structured inputs in a specific format (e.g., if-then rules, form
fields, scripts, queries).
In other words, Non-LLM systems do not generate creative
responses; they only execute the predefined operations assigned to
them.
Type of Non-LLM Tool Prompt Type Example Input Command

Rule-based expert systems Logical rules IF temperature > 50 THEN fan = on

Machine learning classifiers Labeled data + parameter settings predict(class, new_input=[2.3, 1.1, 0.5])

Knowledge-based search engines Query language SELECT * FROM products WHERE price < 100

Image processing algorithms Operation parameters process_image("[Link]", filter="gaussian")

Robotics control software Command chains MOVE X 10; GRAB; MOVE Y 5

Feature LLM Non-LLM

Input format Free natural language Structured commands

Flexibility Very high (creative) Low (strict rules)

Response generation Predictive text generation Rule-based execution

Need for prompt engineering Very high Limited / rarely needed

Engineering Field Example Tool Command (Prompt) Example Expected Output
Electrical and Electronics
SPICE .tran 0.1ms 20ms Runs transient simulation for 20 ms
Engineering
Performs AC frequency sweep 1 kHz–
.ac dec 10 1k 1Meg
1 MHz
Mechanical Engineering ANSYS MP,EX,1,210000 Assigns 210 GPa elastic modulus
F,12,FY,-500 Applies -500 N force on node 12
Industrial Engineering Arena Simulation PROCESS: StationA DELAY 5 MIN Defines 5-minute process time
RESOURCE: Worker CAPACITY 3 Creates a 3-worker resource
Compiles C code and generates
Computer Engineering Compiler / IDE gcc main.c -o main
executable
pytest test_api.py -v Runs all unit tests in test_api
Software Engineering CI/CD Pipeline mvn clean install Builds the project with Maven
Creates a Docker image named
docker build -t myapp .
myapp
Creates a beam between given
Civil Engineering SAP2000 / ETABS FRAME ADD 1 0 0 0 5 0 0
coordinates
ANALYZE RUN Starts the structural analysis
Architecture AutoCAD LINE 0,0 10,0 Draws a line from (0,0) to (10,0)
LAYER SET Walls Switches to the “Walls” layer
Industrial Design SolidWorks EXTRUDE Boss-Base 50mm Creates 50 mm thick solid extrusion
ASSIGN MATERIAL: ABS Applies ABS material to part
Applies Gaussian blur to selected
Visual Communication Design Adobe Photoshop Filter > Blur > Gaussian Blur
layer
File > Export > PNG Exports current design as PNG
Interior Architecture &
SketchUp Line 0,0 to 0,5 Draws a 5 m wall
Environmental Design
Paint Bucket > Material Applies chosen material texture
SELECT * FROM Parcels WHERE Area
Urban and Regional Planning ArcGIS Selects parcels larger than 500 m²
> 500
PROJECT from=WGS84 to=UTM36 Converts coordinate system
Purpose Command Example Description

Define voltage source V1 in 0 DC 5 5 V DC source

Start transient analysis .tran 0.1ms 20ms Time analysis for 20 ms

AC analysis .ac dec 10 1k 1Meg Frequency sweep from 1 kHz to 1 MHz

Save output data .print tran V(out) Prints output node voltage
Purpose Command Example Description

Assign material MP,EX,1,210000 Elastic modulus of 210 GPa

Define load F,12,FY,-500 -500 N force on node 12

Generate mesh AMESH,ALL Mesh the entire model

Start analysis SOLVE Starts the solution process

Purpose Command Example Description

Define processing time PROCESS: StationA DELAY 5 MIN 5 minutes processing time

Set resource capacity RESOURCE: Worker CAPACITY 3 Resource of 3 workers

Start simulation RUN 1000 Runs simulation for 1000 time units

Generate report REPORT UTILIZATION Produces utilization report

Purpose Command Example Description

Compile code gcc main.c -o main Compiles C file and generates executable

Run unit tests pytest test_api.py -v Runs tests for the test_api module

Memory leak test valgrind ./main Checks for memory leaks

Version control commit git commit -m "Added login feature" Saves changes to repository
Purpose Command Example Description

Trigger pipeline git push origin main Pushes new version to main branch

Start test stage npm run test Runs all tests

Build generation mvn clean install Builds project with Maven

Create Docker image docker build -t myapp . Builds a Docker image

Purpose Command Example Description

Define beam FRAME ADD 1 0 0 0 5 0 0 Creates beam using coordinates

Define material MATERIAL ADD CONC f'c=30MPa 30 MPa concrete

Assign load LOAD ADD DEAD 10 kN/m 10 kN/m dead load

Start analysis ANALYZE RUN Starts the analysis

Purpose Command Example Description

Draw line LINE 0,0 10,0 Draws a line from (0,0) to (10,0)

Switch layer LAYER SET Walls Switches to “Walls” layer

Add dimension DIMLINEAR Adds linear dimension

Export to PDF PLOT Prints/export drawing

Purpose Command Example Description

Create new part NEW PART Starts a new part design

Create solid body EXTRUDE Boss-Base 50mm Creates 50 mm thick solid

Assign material ASSIGN MATERIAL: ABS Assigns ABS material

Create assembly NEW ASSEMBLY Starts an assembly

Purpose Command Example Description

Create layer Layer > New > Layer Creates a new layer

Apply filter Filter > Blur > Gaussian Blur Applies Gaussian blur

Change color profile Edit > Color Settings Changes color space

Export to PNG File > Export > PNG Exports as PNG

Purpose Command Example Description

Draw wall Line 0,0 to 0,5 Draws 5 m wall

Copy object Move + Ctrl Moves object while copying

Add lighting Window > Shadows Adds lighting/shadows

Apply material Paint Bucket > Material Applies material

Purpose Command Example Description

Map query SELECT * FROM Parcels WHERE Area > 500 Selects parcels > 500 m²

Add layer Add Layer > [Link] Adds road layer

Coordinate transformation PROJECT from=WGS84 to=UTM36 Converts coordinate system

Export map Export Map as PDF Saves map as PDF

Electrical and Electronics Engineering
LLM Tools:
• GPT-4, Claude, Code Llama → Generating Python/Matlab code for
circuit analysis.
• Gemini → Explaining signal processing algorithms.
Non-LLM Tools:
• MATLAB Signal Processing Toolbox, LabVIEW → Signal/noise
filtering.
• SPICE, Multisim → Circuit simulation.
• TensorFlow (CNN, RNN) → Signal/image classification.
Mechanical Engineering
LLM Tools:
• ChatGPT, Code Llama → Solving thermodynamic equations,
generating FEM code.
• GPT-4 + MATLAB → Scripts for mechanical system analysis.
Non-LLM Tools:
• ANSYS, COMSOL Multiphysics → Fluid dynamics, thermal analysis.
• SolidWorks Simulation → CAD-based structural analysis.
• OpenFOAM → Computational fluid dynamics (CFD).
Industrial Engineering
LLM Tools:
• Claude, GPT-4 → Writing production planning reports, explaining
inventory optimization.
• Gemini → Scenario generation for decision support.
Non-LLM Tools:
• Gurobi, CPLEX, Lingo → Mathematical optimization.
• Arena Simulation, AnyLogic → Process simulation.
• RapidMiner → Data analytics.
Computer Engineering
LLM Tools:
• GPT-4, DeepSeek-Coder, Code Llama → Algorithm development,
debugging.
• Gemini → Documentation for system designs.
Non-LLM Tools:
• Scikit-learn, TensorFlow, PyTorch → Machine learning applications.
• Wireshark → Network analysis.
• VMware, Docker → System testing and simulation.
Software Engineering
LLM Tools:
• GitHub Copilot, Code Llama, ChatGPT → Assisting software
development.
• Claude → API documentation.
Non-LLM Tools:
• Jenkins, SonarQube → Continuous integration and testing.
• Postman → API testing.
• Kubernetes → Application deployment.
Civil Engineering
LLM Tools:
• GPT-4, Claude → Formula generation for concrete mix designs.
• Gemini → BIM (Building Information Modeling) explanations.
Non-LLM Tools:
• ETABS, SAP2000 → Structural analysis.
• [Link] → Bridge and building design.
• AutoCAD Civil 3D → Infrastructure planning.
Architecture
LLM Tools:
• GPT-4 + MidJourney → Concept generation and visualizations.
• Claude → Material selection reports.
Non-LLM Tools:
• AutoCAD, Revit, Rhino → Architectural modeling.
• SketchUp → 3D visualization.
• EnergyPlus → Energy efficiency simulations.
Industrial Design
LLM Tools:
• ChatGPT, Claude → User experience scenarios.
• GPT-4 + Stable Diffusion → Product sketch generation.
Non-LLM Tools:
• CATIA, SolidWorks → Product design.
• KeyShot → Photorealistic rendering.
• ABAQUS → Product durability simulations.
Visual Communication Design
LLM Tools:
• GPT-4, Gemini → Advertising scriptwriting.
• ChatGPT + DALL·E → Poster and logo design.
Non-LLM Tools:
• Adobe Photoshop, Illustrator → Graphic design.
• After Effects, Blender → Animation and 3D modeling.
• Figma → UI/UX design.
Interior Architecture and Environmental Design
LLM Tools:
• ChatGPT, GPT-4 → Conceptual design narratives.
• Claude + MidJourney → Interior visualization proposals.
Non-LLM Tools:
• 3ds Max, Lumion, V-Ray → Interior rendering.
• Ecotect Analysis → Environmental design simulation.
• AutoCAD Architecture → Detailed drawings.
Urban and Regional Planning
LLM Tools:
• GPT-4, Claude → Urban planning reports.
• Gemini → Sustainable planning scenarios.
Non-LLM Tools:
• ArcGIS, QGIS → GIS analysis.
• UrbanFootprint → City simulation, population projection.
• MATSim → Transportation network modeling.
Electrical and Electronics Engineering

Prompt:
“Write Python code to calculate the time constant of an RC circuit.”
LLM Output:
GPT-4 generates working Python code with explanations.
Non-LLM Output:
MATLAB/Simulink computes results via built-in functions, simulation
graphs from SPICE.
Mechanical Engineering

Prompt:
“Generate MATLAB code for analyzing a piston engine’s thermal
efficiency.”
LLM Output:
ChatGPT provides MATLAB code + explanation of thermodynamic
formulas.
Non-LLM Output:
ANSYS/COMSOL performs numerical simulations with detailed
thermodynamic maps.
Industrial Engineering

Prompt:
“Explain the EOQ model and solve an example with D=500, K=100,
h=2.”
LLM Output:
Claude calculates EOQ and explains assumptions in text.
Non-LLM Output:
Gurobi or CPLEX solves EOQ optimization through mathematical
programming.
Computer Engineering

Prompt:
“Write Python code for Dijkstra’s algorithm.”
LLM Output:
GPT-4 produces complete code with comments.
Non-LLM Output:
Scikit-learn or NetworkX runs optimized graph algorithms on datasets.
Software Engineering

Prompt:
“Document an API for a weather forecast system.”
LLM Output:
Claude produces structured API documentation with examples.
Non-LLM Output:
Postman tests the API endpoints directly, Jenkins integrates testing
into pipelines.
Civil Engineering

Prompt:
“Calculate the material mix for C25/30 concrete.”
LLM Output:
GPT-4 explains proportions and generates formula.
Non-LLM Output:
ETABS or SAP2000 analyzes how this mix performs structurally in
simulations.
Architecture

Prompt:
“Design a sustainable wooden house concept.”
LLM Output:
GPT-4 provides conceptual description, MidJourney creates
visualization.
Non-LLM Output:
Revit/SketchUp generates technical CAD drawings and energy
simulation.
Industrial Design

Prompt:
“Generate design ideas for an ergonomic office chair.”
LLM Output:
GPT-4 suggests user scenarios, Stable Diffusion creates sketches.
Non-LLM Output:
CATIA/SolidWorks produces detailed 3D models, KeyShot renders
visuals.
Visual Communication Design

Prompt:
“Create a poster for a sustainable energy campaign.”
LLM Output:
ChatGPT generates text slogans, DALL·E provides poster image.
Non-LLM Output:
Adobe Illustrator produces high-resolution graphic design, After
Effects animates.
Interior Architecture and Environmental Design

Prompt:
“Design a minimalistic living room concept with natural light.”
LLM Output:
Claude provides layout description, MidJourney renders sample
images.
Non-LLM Output:
3ds Max/Lumion create full 3D models with lighting analysis.
Urban and Regional Planning

Prompt:
“Plan a smart city district with renewable energy integration.”
LLM Output:
GPT-4 provides written planning scenarios and sustainability metrics.
Non-LLM Output:
ArcGIS maps land usage, MATSim simulates transportation flows.
[Link] KESİM

0 533 7693205
mkesim@[Link]
[Link]@[Link]
[Link]
[Link]
UIeCaQ
[Link]
[Link]