Why it works

The gap in the market. The science behind the system. The proof it holds up.

Every other player monitors, documents, or digitises. AssetBlue is the reasoning layer the industry has been missing — grounded in an authoritative industrial corpus and validated against reality.

The landscape

The pieces exist. The integrated system does not.

A $17B market of point solutions. None combines structured causal reasoning, a curated failure knowledge base, and zero-infrastructure entry. AssetBlue sits in the empty quadrant.

1

Predictive Maintenance & APM

Enterprise EAM and anomaly detection over telemetry. $500K–$2M+ entry, 12–24 month deployment, requires sensors and historians.

Gap: No diagnostic reasoning. Tells you when, not why.
2

RCA Documentation Software

Templates for recording Fishbone, 5 Whys, and fault trees. Some AI-assisted auto-fill, but the human drives every reasoning step.

Gap: Documentation shells. No domain knowledge built in.
3

Service Resolution AI

Conversational troubleshooting over enterprise tickets and manuals. Workflow-based fix playbooks, retrieval-based not causal.

Gap: Can't generalise to novel failures or explain why.
4

Field Ops / Workflow Digitisation

Digitised SOPs, inspection checklists, and mobile workflows for deskless teams. Execution platforms, not reasoning platforms.

Gap: Digitises known workflows. Doesn't reason about unknown ones.
The market paradox

The plants that need diagnostic intelligence most — under-instrumented facilities with retiring workforces and no CMMS — are the ones that can never afford enterprise APM.

The entire market is designed for the top 5% of facilities. AssetBlue is the first platform for the other 95%.

The moat

What's easy to copy vs. what's hard to replicate.

EASY TO COPY
  • ·Conversational UI wrapper around an LLM
  • ·Basic RAG pipeline over technical PDFs
  • ·Generic troubleshooting chatbot
  • ·Mobile form-factor for field teams
HARD TO REPLICATE
  • Domain-specialised SLM trained on failure ontology, not general web data
  • Multimodal reasoning across text, voice, image, and video in field conditions
  • Curated, semantically indexed failure corpus per asset class
  • Shared reasoning layer across RCA, compliance, and asset life
  • Coaching-first interaction calibrated to engineer expertise
  • Field-validated across real plant conditions and languages

Anyone can wrap an LLM. Few can assemble the corpus, field-validate across plant conditions, and compound knowledge one asset class at a time. That's the moat.

The science

Built from authoritative industrial sources.

The knowledge base is curated from OEM maintenance manuals, IEEE and ASME technical standards, EPRI and VGB PowerTech industry research, and 171 real failure case studies. Every diagnostic output traces to a specific document in this corpus. Nothing is generated from general training data.

Source-grounded, not hallucinated

Qualitative first. Probabilistic over time.

AssetBlue starts by capturing engineering reasoning before enough failure data exists. As evidence accumulates, the system becomes increasingly quantitative, probabilistic, and self-improving.

01

Knowledge-First

LLM + Causal Graph

Extract causal structure from manuals, engineering text, prior cases, and domain knowledge. Use Ishikawa, FTA, and related reasoning frameworks as scaffolds for structured investigation.

02

Prior Elicitation

LLM as Expert Proxy

Structured prompting to generate initial priors over likely relationships, mechanisms, and hypotheses. Treat these as starting beliefs, not ground truth.

03

Bayesian Learning

Probabilistic Ranking with Uncertainty

As case data accumulates, refine priors with observed evidence and transition toward data-driven probabilistic inference and confidence-weighted hypothesis ranking.

Cognitive architecture

Five layers. One auditable reasoning loop.

Perceive

Text, photo, video, voice, operating evidence

Understand

Retrieve context, history, manuals, and prior cases

Reason

Maintain hypotheses, evidence, eliminated causes, and confidence

Coach

Adapt to the engineer's experience and explain the logic

Improve

Every case updates the knowledge base and future reasoning

Every investigation updates an auditable state: symptoms, hypotheses, evidence, eliminated causes, confidence, and next best question.

Validation

Live on the first asset class. Measured against reality.

Automated evaluation across all 171 expert-validated failure cases spanning 10 industries. Four dimensions scored by an independent LLM judge against expert ground truth.

+0%
Lift over plain-LLM baseline on the hardest cases
0%
Root cause accuracy (KB-augmented)
0%
Failure mechanism match (KB-augmented)
0%
Cases where knowledge base wins or ties
Corpus scale
208,773
Document chunks indexed
38,250
Diagnostic images classified
171
Structured case studies
47
Failure mechanisms
220
Operating envelopes
57
Asset component nodes
Evaluated across
UtilitySteelFood ProcessingPulp & PaperChemicalPharmaceuticalOil & GasMiningBuilding ServicesCustom Engineering
Where the KB matters most

On the hardest cases — unusual mechanisms, cross-industry failures, sparse prior documentation — the knowledge base turns misses into diagnoses.

Hardest cases — accuracy lift
Baseline: 0.57
KB: 0.95
Ground-validated with Karnataka Power Corporation (KPCL).
Our standard
>0%
RCA Correctness Rate. Per asset class. Per plant.

Not a target — a floor. A wrong diagnosis acted on once means the engineer never opens the app again. We validate against physical outcomes, not satisfaction surveys.

Correctness over speed

A fast wrong answer destroys trust permanently.

Cited, never asserted

Every output traces to a case study, standard, or validated pattern.

Human-reviewed improvement

Nothing deploys without expert sign-off.

60-day recurrence tracking

If the same failure recurs, the diagnosis was wrong — and the system learns.

Founding research team

Six researchers across computer vision, NLP, edge AI, and applied ML — spanning IIIT Hyderabad and industry.

6
Researchers
4
Domains
48,000+
Combined citations
20+
Patents
Research Domains
Computer Vision
Natural Language Processing
Edge AI & Inference
Applied Machine Learning
Selecting pilot partners — power, oil & gas, manufacturing, mining & minerals

Describe the problem.
Let's reason through it together.

Request a demo. We'll walk you through a live diagnostic session in 30 minutes. Or request a pilot implementation with your asset class.

Or email us at info@assetblue.ai

From the team that built Embibe — India's largest AI education platform, backed by Reliance Jio.