Scattered evidence
Measurements, alarms and maintenance records sit in separate systems. Reconstructing what actually happened means switching between tools, exports and spreadsheets.
AI for industrial diagnostics
Find the root cause,
not just the anomaly.
deepRCA is an agentic AI system for root cause analysis in industrial systems. It brings measurements, alarms, maintenance records and engineering knowledge into one investigation, tests competing explanations, and returns a diagnosis that is backed by evidence.
StageProof of concept
DomainIndustrial systems
First testbedWater distribution networks
01 · The core problem
Detecting an anomaly does not explain its cause.
Monitoring systems are good at raising alarms. The hard part comes next: engineers must connect the evidence, rule out alternative explanations and decide what to do, usually under time pressure and with incomplete data.
Ambiguous symptoms
Different faults and operating changes can produce similar observations. A pressure drop may be a leak, a valve, a pump or simply a change in demand.
Time-consuming investigation
Engineers must bring the evidence together and test possible explanations, often relying on a few experienced specialists whose knowledge is hard to scale.
The challenge is to turn scattered evidence into a supported diagnosis.
02 · How it works
Agentic AI System that runs the investigation with you.
At the center of deepRCA is an LLM agent that coordinates a multi-step investigation. It does not guess from text alone: it calls analytical tools, queries data sources and follows engineering procedures, then reports what the evidence supports.
Agent architectureRunning
- 01
Take in the incident
An alarm or an engineer's question starts the investigation, together with the context of the affected assets.
- 02
Form hypotheses
Using skills and domain knowledge, the agent lists plausible causes, not just the most obvious one.
- 03
Gather evidence
It calls tools and queries data sources through APIs and MCP: time series, events, documentation, models.
- 04
Verify and rule out
Hypotheses are checked with analytical and simulation tools. Explanations the evidence contradicts are dropped.
- 05
Report findings
A conclusion linked to its supporting evidence, with the remaining uncertainty stated openly.
03 · The deepRCA approach
Specific to your system. Verifiable against evidence.
General-purpose AI assistants can sound convincing without being right. deepRCA combines agent capabilities with four diagnostic principles that keep every investigation grounded.
System context
Interpret observations using equipment relationships, operating conditions and domain knowledge. The same reading means different things in different parts of a plant.
Topology-aware reasoning
Competing explanations
Compare alternative causes and actively seek evidence that could rule them out, instead of confirming the first idea that fits.
Hypothesis elimination
Analytical verification
Check hypotheses using analytical tools and available physical or simulation models. Numbers are computed, not generated.
Models over guesses
Traceable conclusions
Link every conclusion to its supporting evidence and make the remaining uncertainty explicit, so engineers can review and trust the result.
Audit-ready output
The proof of concept tests whether these principles improve diagnostic performance.
Inside an investigation
Every step is visible.
deepRCA does not return a black-box answer. Each investigation is a trace of hypotheses, tool calls and evidence that an engineer can follow, question and reuse.
Illustrative trace · simulated water network scenario
04 · Proof of concept
Testing diagnosis in a dynamic, interconnected system.
We start with water distribution networks: a realistic, interconnected infrastructure where faults propagate through the system and symptoms rarely appear where the cause is.
Water distribution network · simulationMonitoring
Why a water network?
Controlled ground truth
Simulation provides controlled fault scenarios and known root causes, so diagnostic quality can be measured objectively.
Current work
Scenarios & data
Preparing fault scenarios and monitoring data for diagnostic experiments: leaks, valve and pump faults, sensor issues and demand changes.
What we aim to validate
Evidence-backed RCA
Root cause identification against selected baselines, with evidence supporting each conclusion.
Now
Proof of concept
Simulated fault scenarios in water networks. Focus on diagnostic accuracy, evidence quality and limitations.
Next
Pilot deployments
Field validation with partners on real operational data and real incidents.
Later
Industrial systems
Extending the approach to other interconnected industrial and critical infrastructure systems.
05 · About us
The team behind deepRCA.
Research in automation, robotics and complex systems, combined with experience building production AI and petabyte-scale data platforms.
CEO
Adam Borowa, PhD
Project Leader
AI engineer and researcher with a PhD in Automation and Robotics and experience in LLM-based systems and AI agents. At deepRCA, he oversees the project's technical development, coordinates fault simulation and experimental activities, contributes to AI agent architecture, and ensures the integration of individual components into a coherent diagnostic system.
CTO
Grzegorz Reinelt
AI Technology Lead
Drives deepRCA's AI engineering and technology architecture, bringing extensive experience in building production-grade AI platforms and scalable software systems. Specializes in agentic AI, LLM-based reasoning, multi-agent orchestration, and automated evaluation, transforming advanced AI capabilities into robust, scalable diagnostic intelligence.
CDO
Grzegorz Ewald
Data Architecture Lead
Data architect with experience designing petabyte-scale data platforms, specializing in data modeling, data lakes, and cloud-based architectures. His research background spans distributed and multi-agent systems, mathematical modeling, simulation, and optimization, with a particular focus on critical infrastructure. At deepRCA, he brings together practical experience in large-scale data systems and a scientific understanding of complex system behavior to support the development of intelligent diagnostic solutions.
Work with us
Help us take deepRCA from simulation to the field.
We are looking for engineering teams and organisations that want to shape the next generation of diagnostic tools for industrial systems.
Domain expertise
Pilot access to operational data
Industry partnerships
Funding