Skip to content

AI-led yield optimization

Predict Yield.
Optimize Performance.

Every day your hydrocracker runs, a decision gets made with only part of the picture, and the margin from the rest of it disappears quietly.

Refinery A · 31 months of one hydrocracker · middle distillate, bbl/day
See how it works ↓Live demo ↗
Left on the table, this run
$0.0M
  1. 1 / 6

    This is one hydrocracker, every day for 31 months.

    Middle distillate output from a real, anonymised unit across a full catalyst run: 881 operating days of history.

  2. 2 / 6

    Some weeks it runs far better than others.

    The band is what this unit typically makes at each point in the catalyst's life. The dashed line is its best demonstrated weeks: not a design figure, something it has already done.

  3. 3 / 6

    That spread is money.

    Valued at each month's middle-distillate crack over unconverted oil, the gap to its own best weeks comes to $53.5M over the run. About $22M a year.1

  4. 4 / 6

    Silos Cost You Yield

    Operations, logistics and commercial each optimize their own piece. None of them sees the whole one.

    • Operations: what they see. Unit performance, not what logistics needs or what a contract is worth.
    • Logistics: what they see. Vessel and shipping constraints, not how a feed change shifts yield.
    • Commercial: what they see. Pricing and contracts, not the operational cost of hitting them.

    Decisions made in isolation. Margin left on the table.

  5. 5 / 6

    Operations' piece comes down to two temperatures.

    R1 strips nitrogen out of the feed; R2 does the cracking. Run R1 hotter and less nitrogen slips through, so R2 needs less heat for the same conversion. Move R1 and see what happens.

  6. 6 / 6

    Operators know which way to move. The model tells them how far.

    Each month it reads the feed, knows how this unit responds at this point in the catalyst's life, and proposes R1 and R2 together. Here: R1 up 2.7 °C and R2 down 0.7 °C, holding conversion. Your engineers accept, defer or reject.

Now try it yourself.

What your engineer sees: the option, its predicted yield impact, and how confident the model is. What happens on approval: the change is logged, and the outcome is tracked against the prediction.

Open the simulator →
The full simulator · runs in your browser · best on a laptop or larger
hydrocracker.net/demo
The simulator's choice cascade: feed, R1, nitrogen slip, R2, conversion and a per-day dollar ledger.

Embedded, Not Consulted

Our engineers work inside your operation, not from a slide deck.
  1. Step 1
    NDA

    Before anything else.

  2. Step 2
    Diagnostic session

    We learn your unit, your data, your constraints.

  3. Step 3
    3–4 week POC

    A working model, on your data, reviewed by your team.

From Data to a Decision You Can Act On

We connect what you already have (historian data, planning models, contracts, shipping constraints) into one model of your unit. It predicts the yield impact of a decision before it's made, and ranks the options. Your engineers review every recommendation before anything changes. Nothing acts on its own.

Connect your data
Read-only access, no changes to your infrastructure.
Predict the outcome
Trained on your unit's own operating history, not a generic model.
Review before acting
Every recommendation goes to your engineers first.

Why Hydrocracker.ai

  • Unlike LP and RTO models, it updates continuously, not quarterly.
  • Unlike consulting, you see a working model in 3–4 weeks, not after months of scoping, and we stay embedded with your team.
  • Unlike building it yourself, it doesn't depend on who happens to be on staff.
  • Unlike any of them, it sees operations, planning and commercial in one view, and we're accountable for the outcome alongside you.

Read-only access. No changes to your infrastructure. Every recommendation reviewed by your engineers before anything changes.

Who We've Worked With

“They tied together data sources that had never talked to each other before, and gave us real-time visibility into our trading decisions and the margin impact behind them. That turned into a strategic advantage, not just an operational one.”

Regional Director, Large US-based export refinery

“What stood out was the review step — every recommendation came with the reasoning, and our people made the call, not the model. And they moved fast: we had something real in front of us in weeks, not a quarter of scoping meetings.”

GM Refinery Operations, State-owned Refinery

Shared with permission. Names are withheld to respect client confidentiality.

Who We Are

Rahul (راهول)

Founder & CEO

$1B+ in AI-led value delivered across enterprise engagements spanning AI strategy, commercial optimization, and oil & gas supply chain. Leads client relationships and company direction.

Chandra (تشاندرا)

Senior Advisor, Refining & Process Engineering

35 years heading refining operations at state-owned South Asian refineries. Expert in hydrocracking chemistry, catalysts, and yield optimization. Has run hydrocracker units at scale and knows every lever from the inside.

Claude (كلود)

AI Reasoning Engine, powered by Anthropic

Frontier AI embedded as Hydrocracker.ai's core reasoning layer, holding chemistry, economics, and live market data in one model. Runs in advisory mode: every output is reviewed by our team before anything changes.

Let's look at your unit.

A 3–4 week POC costs far less than the margin you're already losing.

Try it first
Open the demo ↗