Assistant
Inform Online 2022 – Unleash the Potential of Your Data with Access Inform (TotalEnergies)
How a Global Energy Leader Built MILAa & JAFAR to Cut Refinery Downtime
Unplanned downtime in the oil and gas industry costs offshore organizations an average of $38 million annually — and for the worst performers, financial impacts can exceed $88 million per year (Kimberlite). In refinery and processing operations, a single unplanned outage runs approximately $125,000 per hour. The root cause is rarely a technical failure alone. It is a knowledge failure: the inability to quickly locate the root cause analysis from a prior incident, surface the maintenance record for a specific piece of equipment, or cross-reference failure patterns across sites and teams before they repeat.
TotalEnergies — one of the world’s five largest energy companies — built two AI-powered knowledge assistants on Sinequa’s enterprise search platform to solve this problem at their refinery, a site with close to 1,700 employees. In this Inform Online session, Pierre Jallais, Lead Architect for Smart Search Engines and LLMs at TotalEnergies, presents how TotalEnergies built MILAa and JAFAR — and what those systems now make possible for operational engineers.
Watch the video
The Three Operational Problems MILAa and JAFAR Solve
Before this deployment, TotalEnergies’ engineers faced three compounding risks from fragmented operational knowledge:
- Extended downtime — Root cause identification was delayed because relevant RCA data could not be found quickly enough when failures occurred. At $125,000 per hour of unplanned downtime, every hour of delayed diagnosis has a measurable financial cost.
- Inconsistent decision-making — Operational learnings were not shared systematically across teams or sites. Engineers made decisions without access to the full body of organizational experience, leading to inconsistent problem resolution and missed opportunities to apply what had already been learned.
- Repeated incidents — Without accessible RCA data, the same failures recurred. The pump failure example — multiple incidents over two years that could have been prevented — illustrates the cost of knowledge that exists in documents but cannot be found when it matters.
MILAa and JAFAR directly address all three. By making 1,000+ RCA documents queryable in natural language across four languages, they compress root cause identification from hours to minutes, standardize operational responses across TotalEnergies’ global sites, and create a self-reinforcing knowledge base that makes every new incident finding available to every engineer immediately.
Why This Session Matters for Energy Sector AI Leaders
This Inform Online 2022 session captures the founding vision behind what became one of the most documented industrial AI deployments in the energy sector. The strategy Pierre Jallais outlines — using Sinequa as the intelligent search foundation, building domain-specific ontologies for industrial vocabulary, deploying multilingual NLP for a globally distributed workforce, and constructing AI assistants that answer in natural language from vetted operational data — is the same architecture that energy organizations worldwide are now trying to replicate.
The deployment was the first TotalEnergies site to leverage AI-powered search for RCA and incident analysis. The approach has since been recognized as a model for how industrial organizations can move from fragmented document management to connected, conversational operational intelligence — without replacing source systems, without requiring engineers to become data scientists, and without accepting the hallucination risk that comes with generic GenAI tools disconnected from proprietary operational data.
Frequently Asked Questions (FAQ)
MILAa (My Intelligent Learning Application) and JAFAR (Generative AI for Return of Experience) are two AI-powered knowledge assistants built by TotalEnergies on Sinequa’s enterprise search platform at their Antwerp refinery. MILAa consolidates more than 1,000 Root Cause Analysis documents into a centralized, AI-enhanced knowledge base, allowing engineers to query equipment failure histories, downtime durations, and corrective actions in natural language. JAFAR extends this with a generative AI conversational layer that translates technical RCA documents into French, Dutch, German, and English in real time, using a custom-built internal dictionary to preserve TotalEnergies’ domain-specific terminology and acronyms. Together, they transform decades of accumulated operational knowledge from inaccessible static documents into queryable, conversational intelligence.
Before deploying MILAa and JAFAR, TotalEnergies’ Antwerp engineers faced three compounding problems from fragmented operational knowledge. Root cause identification was delayed because 1,000+ RCA documents were scattered across siloed systems in inconsistent formats and single languages, meaning engineers manually searched dense reports under time pressure during active incidents. Operational learnings were not shared across teams or sites, leading to inconsistent decisions. And the same failures recurred — in one documented case, multiple pump failures over two years could have been prevented if relevant RCA data had been accessible in real time. Unplanned downtime in oil and gas costs an average of $38M annually (Kimberlite), and approximately $125,000 per hour, making the cost of inaccessible operational knowledge directly quantifiable.
JAFAR automatically translates TotalEnergies’ technical RCA documents into French, Dutch, German, and English while preserving domain-specific industrial terminology and internal company acronyms. This multilingual capability required building a custom internal dictionary integrated with the base language model — ensuring that Sinequa understands TotalEnergies’ specific vocabulary rather than applying generic industrial language approximations. For a global energy company operating refineries across multiple countries and language environments, this is a critical capability: engineers across sites can query operational knowledge in their native language and receive answers that reflect TotalEnergies’ actual terminology, not a translated approximation that loses technical precision.
Share
