Back to Blog

How We Built a Production OS That Increased Manufacturing Efficiency by 30%

By B.S. Karthigeyan · January 2026 · 12 min read

When KRG Medifabb Pvt Ltd — a Chennai-based medical disposables manufacturer — approached us in November 2025, their operations were drowning in paper. Attendance in registers. Inventory in notebooks. Production output in Excel files that were always a day out of date. Payroll took the HR team nearly three full days every month.

Six weeks later, we handed them a fully live Production Operating System that cut those three days of payroll work down to under three hours, gave management a real-time view of every production line, and reduced material waste by 10%. Here's exactly how we built it.

Starting with the problems, not the solution

The single biggest mistake in custom software is jumping to the solution before deeply understanding the problem. We spent the first two weeks doing nothing but watching how the factory actually ran.

We shadowed the floor supervisors. We sat with HR during payroll week. We interviewed the operations director about what decisions he was making blind. Three clear problems emerged — payroll was broken, production was invisible, and inventory was a constant source of waste and surprises.

The architecture decision that changed everything

We almost chose a cloud database. The conventional choice would have been PostgreSQL on AWS RDS — scalable, reliable, industry-standard. But after our discovery, we learned something critical: the factory floor had intermittent internet. Workers would need to log data even when the connection dropped.

We chose SQLite with a sync layer instead. This meant the system worked 100% offline on the factory tablets, and synced to the central dashboard whenever connectivity was restored. A boring, simple choice that saved the entire project.

The full stack: React 18 for the frontend (fast, component-based, great for dashboards), FastAPI on Python for the API layer (Python's data handling made waste calculation formulas trivial to implement), SQLite for on-device persistence, and Vercel for deployment.

Building the four modules

The system was divided into four integrated modules, each designed around a specific job-to-be-done that we had identified during discovery.

Production Dashboard: A live view of every production line, updated every 5 minutes. Shift-wise output comparisons, daily vs. target tracking, and automated alerts when waste percentage crossed thresholds. Management could now see the factory from their phone.

HR & Payroll Module: Digital attendance with supervisor sign-off, automated overtime calculation using the factory's exact piecework formulas, deduction management, and one-click salary generation. Payroll dropped from 3 days to under 3 hours.

Inventory Control: Every raw material item had a live stock count. The system automatically calculated consumption based on production entries and flagged reorder points. Overordering and stockouts were essentially eliminated.

Automated Reports: A one-click daily summary report, weekly performance digest, and monthly production data in a format that fed directly into the management's existing Excel-based P&L. No more manual report compilation.

The results after 60 days

Sixty days after go-live, we sat down with the KRG Medifabb team to review the numbers. Every metric had moved in the right direction.

Overall production efficiency was up 30% — not because workers were working harder, but because supervisors had real-time data to make faster decisions and management could identify underperforming lines before they became serious problems. Payroll was down to under 3 hours. Material waste had dropped by 10%. And the migrated 100+ employee records showed 100% data integrity after the move.

What we learned

The biggest lesson was the offline-first decision. If we had defaulted to a cloud database without asking the right questions during discovery, the entire system would have been unusable on the factory floor.

The second lesson was scope discipline. We were asked mid-project to add a customer order tracking module. We declined — not because we couldn't build it, but because adding scope mid-build is the fastest way to miss deadlines and dilute quality. We scoped it as a Phase 2 instead, which is exactly how it should be done.