01 · The Problem
A 60-year-old method.
A 2-hour bottleneck.
YBL's quality control workflow relied on Single Radial Immunodiffusion (SRID), a methodology developed in the 1960s.
Scientists manually measured precipitin ring diameters using rulers, entered values into spreadsheets, plotted standard curves by hand, and routed results through a multi-scientist verification chain.
The science was reliable.
The process was not.
What made this urgent
A direct competitor was already using an earlier version of SUN BIO IT's image analysis software. YBL needed to move — and came with a specific ask: make this faster and more accurate without compromising compliance.
🔬Problem context — SRID process or lab environmentThe before: manual measurement, precipitin rings, lab context
02 · My Role
End-to-end ownership —
from first call to production.
I owned the project end-to-end:
- Discovery sessions with YBL's head of QC
- Requirements translation into technical specification
- Full UX design — login, dashboard, scan canvas, report generation
- Developer coordination throughout the build
- Precision calibration testing to resolve accuracy issues
- Client communication and demo cycles through to sign-off
📋Discovery / requirements artefactsResearch notes, wireframes, or client session outputs
03 · The Process
The biggest challenge
wasn't the interface. It was the accuracy.
YBL used images from densitometers, stained gel sheets, phone cameras, and document scanners. Each source introduced different DPI and PPI values, causing inconsistent measurements across the same sample set.
To isolate the issue, I built a set of precision calibration illustrations in Adobe Illustrator — circles and ellipses at exactly known dimensions — and ran them through the software pipeline. The discrepancy was immediate: a conversion error inside the measurement calculation logic.
Once corrected, accuracy stabilised.
A second issue: SRID rings aren't perfect circles. Biological variation makes them slightly elliptical. The original model only detected circles. I worked with the development team to extend detection to ellipses — improving both detection rate and measurement precision.
Accuracy targets — non-negotiable
99% measurement accuracy. 90%+ automatic well detection. Both required for clinical-grade use. Both met.
📐Calibration illustrations — Adobe IllustratorPrecision circles and ellipses used to isolate the DPI error
📊Accuracy validation outputML detection results showing stabilised accuracy
04 · The Solution
Four stages.
Scan to validated report.
The final application guided scientists through four stages:
- Login
- Upload Sample
- Scan & Validate
- Generate Report
The system automatically:
- Detects precipitin rings using the trained ML model
- Measures ring diameters
- Plots the standard curve
- Calculates antigen concentrations
- Exports a validated PDF or Excel report
An ellipse correction tool let scientists handle biological edge cases in seconds — without reprocessing queues or compliance gaps.
Validated under 21 CFR Part 11 — the FDA standard for electronic records in pharmaceutical environments. Designed for compliance from day one.
🖥️Main UI — scan canvas with ring detectionML detection in action, ellipse tool, sample history dashboard
📤Sample upload screenSample ID, concentration values, image upload flow
📄Report generationStandard curve, concentrations, PDF/Excel export
"
A 2-hour process that bottlenecked every batch before release. Now it takes two minutes — and the audit trail is cleaner than it ever was manually.
— Shridevi S S
05 · Outcome
2 Hours → 2 Minutes.
Still running today.
0%Reduction in QC
analysis time
0%Measurement accuracy
validated
0%Automated detection
rate
- 21 CFR Part 11 compliant
- Still in production use at Yashraj Biotec today
📈Before / after — process comparison2-hour manual process vs 2-minute automated workflow