Energy AI Assistant

An AI assistant focused on diagnosing production, facilities, and other operational issues for the oil and gas industry.

Comment

⚡ Energy AI Assistant

💡 Inspiration

Diagnosing oil-and-gas production issues often feels like detective work — mountains of data, multiple experts in silos, and precious hours lost just figuring out who should tackle what. We set out to build a system that thinks like an engineering team: fast, collaborative, and always learning.

🤖 What It Does

Energy AI Assistant is a multi-agent AI system built to accelerate diagnostics in upstream operations.

  • Supervisor Agent: Listens to incoming problem statements or data, interprets them, and routes the issue to the right expert.
  • Production Engineer Agent: Focuses on subsurface / well-performance problems (e.g., unexpected drop in production, GOR shifts, IPR deviations).
  • Facilities Engineer Agent: Focuses on surface/infrastructure issues (e.g., separator upsets, slugging, compressor trips, flowline restrictions).
    Agents “talk” to each other when needed, examine data, debate root causes, and give you actionable, explainable recommendations — just like a real engineering round-table.

🛠️ How We Built It

  • Architecture: Designed a routing layer (Supervisor Agent) and two domain-specific agents, orchestrated by a multi-agent controller.
  • Tech stack: Python backend, modular prompt templates and roles, REST API for integration.
  • Training / Prompting: Each agent was primed with domain knowledge (production optimization, flow assurance, operational diagnostics). We fed the system real-world upstream scenarios — e.g., ESP failures, hydrate formation, abnormal separator pressures.
  • Workflow: Data/input → Supervisor Agent routes → Domain Agent(s) diagnose → Combined response delivered. Built for quick iteration and extensibility.

🚧 Challenges We Ran Into

  • Telling the Supervisor Agent when to route to both domain experts (surface + subsurface) versus just one.
  • Avoiding AI “hallucinations” when data was limited or ambiguous.
  • Ensuring agents maintained domain boundaries — so the Production Agent didn’t drift into pure facilities issues (and vice-versa).
  • Crafting responses that were technically accurate and engineer-friendly (concise, clear, actionable).

🏆 Accomplishments We’re Proud Of

  • Delivered a working prototype that simulates multi-disciplinary engineering collaboration — but entirely in software.
  • Showed the system diagnosing complex scenarios in seconds instead of the minutes or hours a human team might take.
  • Built transparent, step-by-step reasoning logs so users trace the “why” behind each recommendation.
  • Demonstrated that AI can augment expert engineers (rather than replace them) — making real-world use cases feasible.

📚 What We Learned

  • Well-defined roles matter: Giving each agent a strict domain (and hand-off logic) made the collaboration much more coherent.
  • Context is king: Agents need strong memory of state, previous actions, and data sources to avoid circular reasoning or redundant tasks.
  • Explainability matters: Even engineers appreciate a clear narrative (“Here’s the cause, here’s how we fixed, here’s what we’ll monitor”). AI tools that don’t explain are black boxes and lose trust.

🚀 What’s Next for Energy Diagnostic AI Assistant

  • Integrate with live data streams (SCADA, PI, historian) so diagnostics run in near-real time.
  • Add additional domain agents: Corrosion Engineer, Process Safety Engineer, Reliability Engineer.
  • Build a visual dashboard that shows agent interactions, reasoning paths, and recommended actions.

Built With

Share this project:

Updates

Submission history