AlphaIQ mark

AutoSim · An AlphaIQ product

Under the hood.

Two systems do the heavy lifting: a domain-agnostic LLM pipeline that turns any input into structured simulation parameters, and a template system that assembles agents out of tested blocks.

Core technology

Domain-agnostic LLM pipeline

The AI analyses diverse inputs — text, research papers, raw data — and structures the critical simulation components: agent types and behaviours, environmental factors and rules, interactions and relationships.

  1. Input (text, data)
  2. LLM analysis
  3. Structured parameters

Natural language and raw data become precise, runnable simulation parameters.

Universal Agent Template System

UATS is a library of pre-built, highly customisable blueprints for agents across economic, social, biological and physical domains.

  • Enables hybrid agents that reflect real-world complexity
  • Supports sophisticated cross-domain interactions
  • Fosters reusability and accelerates model development
Economic blockSocial blockEnvironmental blockBehaviour blockHybrid agent

Build complex, multi-faceted agents by combining UATS modules.

System workflow

The GenAI ABM pipeline, stage by stage.

  1. Model input

    PDF upload

  2. Data prep

    Calibration & extraction

  3. AI code gen

    Code generation

  4. Simulate

    30 ensemble runs

  5. Validate

    Metrics & assessment

  6. Improve

    Iterative refinement

Model input & understanding

  • PDF upload with frontier-model extraction
  • Model description
  • Key agents identified
  • Parameters suggested

Data preparation & calibration

  • Choose a data source: synthetic generation or CSV upload
  • AI data cleaning
  • Statistics extraction (peaks, trends, magnitudes)
  • Event detection (spikes, drops, shocks)

ABM code generation

  • Structured prompt (fixed + logic + context)
  • PLAN block as the model's scratchpad
  • One-shot examples (multiplicative shocks, clipping)
  • Skeleton template enforcement
  • Multi-dimensional AI review
  • Auto-retry with fixes, up to three attempts

Simulation execution

  • Safety validation (imports, serialisation, JSON)
  • Code downloaded to the browser
  • 30 ensemble Monte-Carlo runs
  • Results collected as means and CI bands

Validation & metrics

  • Smart alignment against calibration columns
  • Ensemble metrics: R², coverage, KS test
  • AI interpretation of results
  • Visual comparison with calibration data

Iterative improvement

  • AI suggests improvements to parameters, logic and structure
  • User approves, or auto-apply
  • Code regenerated with validation context
  • Re-simulate with the new code
  • Re-validate against calibration
  • Compare iterations side by side

Code generation architecture

Section A

  • Import statements
  • JSON output format
  • Serialisation checklist
  • Mandatory structure

Section B

  • One-shot examples
  • Multiplicative shocks
  • Stability clipping
  • Code skeleton

Section C

  • Dynamic context
  • Model description
  • Calibration statistics
  • Event timings & parameters

Code quality assessment

Technical correctness
40%
Code readability
20%
Documentation
20%
Complexity analysis
20%

Validation metrics

  • R² score
  • Coverage rate
  • KS test p-value
  • Peak timing error

Technology stack

AI models
Frontier LLMs
Model specification and code generation
Execution
Pyodide WebAssembly
Real Python, run in the browser
Interface
React + Tailwind
The studio, streamed live as the model is written

Current configuration

Time steps
500
Synthetic steps
200
Ensemble runs
30
Auto-retry max
3
Synthetic runs
5
See the proof
Industry verticals

Next step

Bring a system you need to defend.

Describe it in plain language and AutoSim will build, calibrate and score a simulation of it while you watch. If the question needs more than the product alone, AlphaIQ will build it with you.

Talk to AlphaIQ