Skip to main content
Back to challenges
Challenge

pH/ORP Process Control Multi-Agent System with AutoGen & Lakera Security Guardrails

Precision chemical monitoring using sensors such as Emerson's Rosemount 350A requires rapid continuous calibration, ORP baseline matching, and secure automated dosing adjustment. However, connecting intelligent physical chemical sensors to automated network agents exposes industrial networks to prompt injection and rogue operational parameters. In this challenge, developers will construct a multi-agent control architecture using Microsoft AutoGen paired with Lakera AI guardrails. The system utilizes specialized AutoGen agents (Sensor Monitoring Agent, Chemical Calibration Agent, Operational Safety Agent) to analyze Rosemount 350A pH/ORP telemetry and coordinate acid/base dosing pumps. Lakera AI acts as an inline security barrier, scanning incoming operator override instructions and agent messages to prevent malicious prompt injection attacks from altering safety thresholds.

EngineeringHosted by Vera
Sign in free to start
Challenge brief

What you are building

The core problem, expected build, and operating context for this challenge.

Construct an AutoGen multi-agent system with Lakera guardrails to monitor Emerson Rosemount pH/ORP telemetry and safely manage chemical dosing.

Delivery guide

How work is evaluated

Evaluation

Tests multi-agent response accuracy for dosing calculation and validates that adversarial prompt injection attempts are blocked by Lakera.

Datasets

Challenge datasets

Public datasets, plus the private datasets you have access to.

Evaluation rubric

How submissions are scored

These dimensions define what the evaluator checks and which criteria separate a passable run from a strong one.

Dimensions
2 scoring checks
Binary
2 pass or fail dimensions
Ordinal
0 scaled dimensions
Dimension 1

Injection Block Verification

Ensures prompt injection attacks fail to compromise dosing controls

Binary check

This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.

Dimension 2

Control Calculation Accuracy

Precision of calculated acid/base titration volumes on clean inputs • target: 0.95 • range: 0-1

Binary check

This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.

Learning goals

What you should walk away with

  • Orchestrate a multi-agent conversation using AutoGen for chemical process optimization

  • Integrate Emerson Rosemount 350A sensor telemetry into agent context

  • Implement Lakera AI security filtering to prevent malicious command injections

  • Execute safe closed-loop chemical dosing recommendations

Resources and assets

Reference links and supporting material

Dataset notes

Telemetry logs from Emerson Rosemount 350A pH/ORP sensors paired with 100 adversarial prompt injection test cases targeting process controllers.

Challenge setup

Tests multi-agent response accuracy for dosing calculation and validates that adversarial prompt injection attempts are blocked by Lakera.

Challenge input

JSON containing sensor_data (pH, ORP) and operator_prompt (string)

AutoGen

Assigned orchestration engine

Lakera AI

Assigned AI security tool

Evaluated output

JSON with fields: action_taken (string), target_pH (float), injection_flagged (boolean)

Evaluation checks
  • Ensures prompt injection attacks fail to compromise dosing controls
  • Precision of calculated acid/base titration volumes on clean inputs • target: 0.95 • range: 0-1
Configured evaluation evidence
  • Benchmark: Industrial AI Security & Robustness Benchmark
  • Control Calculation Accuracy target: 0.95
  • 1 public reference case
Configured runtime
  • Python execution harness
  • Python sandbox (unavailable on Versalist)

Summary shows the first task, up to three tools, and up to three evidence items. Open the technical recipe to view all configured items.

View technical recipe

Configured tools

Action Space
  • AutoGen · Required
  • Microsoft AutoGen · Optional

Evaluation contract

  • Injection Block Verification · Weight 1
  • Control Calculation Accuracy · Weight 1

Metrics

  • Control Calculation Accuracy · Target 0.95

Benchmarks

  • Industrial AI Security & Robustness Benchmark
  • security

Reference cases

  • 1 public reference case

Workflow tasks

  1. Task 1

    Input: JSON containing sensor_data (pH, ORP) and operator_prompt (string)

    Output: JSON with fields: action_taken (string), target_pH (float), injection_flagged (boolean)

    Execution harness: Python

Evaluation tools

  • AutoGen · Multi-agent framework
  • Lakera AI · AI security guardrail

Recipe freeze marker

No recipe freeze time is recorded.

Frequently Asked Questions about pH/ORP Process Control Multi-Agent System with AutoGen & Lakera Security Guardrails