Full Stack Intern
Dimensionless Technologies
Built an image-processing pipeline using Node.js, OpenCV.js, and Google Vision API that automated medicine identification from batch photos, replacing manual data entry. Designed an auto-populated Excel-compatible inventory system that reduced medicine cataloging time from hours to minutes per batch.
Role Overview
At Dimensionless Technologies, I worked on automating a manual and time-consuming process in pharmaceutical inventory management. Pharmacies and distributors were cataloging medicines by hand — photographing batches of products, then manually transcribing names, dosages, manufacturers, and expiry dates into spreadsheets. This took hours per batch and was error-prone.
I built an end-to-end pipeline that takes batch photographs of medicines, applies image preprocessing with OpenCV.js (perspective correction, contrast enhancement, noise reduction), extracts text using Google Vision API's OCR, and parses the extracted text into structured inventory fields. The final output auto-populates an Excel-compatible spreadsheet, reducing a multi-hour manual task to a few minutes of automated processing per batch.
Key Responsibilities
Image Preprocessing
Built the image-preprocessing stage using OpenCV.js — perspective correction for skewed photos, adaptive thresholding for varying lighting, contrast enhancement, noise reduction, and automatic region-of-interest detection to isolate individual medicine labels from batch photographs.
OCR Integration
Integrated Google Vision API for optical character recognition, configuring language hints, handling multi-orientation text detection, and implementing confidence-based filtering to discard low-quality extractions before the parsing stage.
Text Parsing & Structuring
Developed regex-based and rule-based parsers to extract structured fields from raw OCR text — medicine names, active ingredients, dosage, manufacturer details, batch numbers, manufacturing and expiry dates — with fuzzy matching for common OCR errors.
Excel Automation
Designed the auto-populated Excel-compatible inventory output using Node.js libraries, generating formatted spreadsheets with medicine details, stock quantities, expiry tracking, and validation highlighting for entries needing manual review.
Batch Upload System
Built the Node.js backend handling batch photo uploads with progress tracking, image-queue management, and sequential processing — allowing users to upload entire batches and receive consolidated inventory reports.
Quality Assurance
Implemented validation and review workflows — flagging low-confidence extractions for manual review, providing side-by-side image-to-text comparison views, and tracking extraction accuracy across different medicine packaging types.
Key Deliverables
Image Processing Pipeline
End-to-end pipeline from raw batch photographs to structured medicine data — OpenCV.js preprocessing, Google Vision API OCR, rule-based text parsing, and structured field extraction with confidence scoring.
Auto-Populated Inventory System
Excel-compatible inventory spreadsheets automatically generated from processed images — medicine names, dosages, manufacturers, batch numbers, expiry dates, and stock quantities — reducing cataloging from hours to minutes per batch.
Batch Processing Interface
Web interface for uploading batch photos with drag-and-drop support, processing progress indicators, side-by-side image-to-extracted-data review panels, and one-click export to formatted Excel files.
Process Automation
Replaced a fully manual data-entry workflow — what previously required hours of transcribing medicine details from photographs was reduced to minutes of automated processing, with human review only for flagged low-confidence entries.