Building our products
Building ASTR Production: What We Learned from Plant Floors
The research, wrong turns and design decisions behind our production AI.
Trikaan Research & Engineering4 min read
KEY TAKEAWAYS
- We expected to build an AI planner; plants needed reliable data capture first.
- Stage-wise yield turned out to be the most valuable early feature.
- AI suggestions are only trusted when they show their sources.
- Approval workflows made adoption possible.
ASTR Production started with an ambitious idea: AI that writes recipes, plans production and generates job cards. Our first plant visits quickly changed the order in which we built it.
This is the story of those decisions.
What we expected
We assumed the hard problem was optimisation — finding the best plan or recipe. We imagined planners feeding requirements to an AI and receiving finished schedules.
What we found
- Production data was captured late, on paper, in different units.
- Yield was known only as a monthly total.
- Recipes lived in spreadsheets with unclear versions.
- Planners’ real constraints were not written anywhere.
Decision 1: Capture before intelligence
We built stage-wise capture of input, output and loss first, designed for tablets on the floor. Within weeks, plants could see where yield was lost — before any AI feature existed.
Decision 2: Recipes as controlled objects
Recipes became versioned records with approvals, substitutions and links to batch results. Only then did AI recipe suggestions make sense: they could learn from which versions performed best with which inputs.
Decision 3: Show the work
Every suggestion shows the batches and data it is based on. Plant heads told us they would not approve a recipe they could not question.
Decision 4: Approval is the product
We stopped treating approvals as a compliance add-on. Drafts, edits, approvals and the reasons behind them became central to the workflow — and the reason teams adopted AI at all.
Frequently asked questions
What does ASTR Production do?
It covers recipes, BOMs, planning, work orders, job cards, quality, costing and maintenance insights, with AI assistance and approvals.
Does it require machine connectivity?
No, but machine data adds OEE and maintenance intelligence.
How does AI learn our process?
From your recipes, batch results and quality data, with outputs reviewed by your team.
Can we start without AI?
Yes — many plants start with capture and recipes, adding AI once data is reliable.
- #AI production planning
- #AI recipe generation
- #manufacturing AI product
- #production AI
- #AI for process manufacturing
Trikaan Research & Engineering
Trikaan
Written by the researchers and engineers who build Anvaya, ASTR and Seed Track, with input from Trikaan's delivery and engineering teams.
