01↗
Computer vision systems
Detection, segmentation, tracking. Built around your data, evaluated on your metrics, deployed to your stack.
pytorchdetectronmmdet
We/Converge
Building intelligent vision systems for real-world impact.
Production-grade AI systems engineered to run at scale. Eight years of ML in production — not in slides.
Converge
02/Services
Things we do, end-to-end.
From a vague problem statement to a system running in production. The middle 80% is where most of the work lives.
02↗
Vision-language models
VLMs for retrieval, captioning, document understanding. Fine-tuned where it matters — not just prompted.
clipsiglipllava
03↗
Visual search & retrieval
End-to-end indexes from photo to product. Index design, ANN, rerankers, eval — the parts most demos skip.
faissqdrantrerank
04↗
OCR & document AI
Invoice, form, receipt and ID extraction. High recall, audited error modes, human-in-the-loop where required.
layoutlmdonutpaddle
05↗
Edge AI & deployment
Quantize, profile, ship. Sub-100ms inference on CPU, GPU and embedded targets.
onnxtensorrtopenvino
03/Selected Work
Systems we've shipped.
A selection of production work across retail, logistics, industrial automation and healthcare. Detection, retrieval, OCR, VLMs — and the unglamorous pipeline work that makes any of it real.
01 · Industrial · SafetyShipped · 2026
Automated monitoring zones at scale
End-to-end pipeline combining monocular depth estimation (Depth Anything V3) and semantic segmentation (OneFormer) to auto-generate per-camera monitoring polygons — enabling people-in-distress detection across venue deployments.
Impact:Auto-generated zones across hundreds of cameras · zero manual annotation
Depth Anything V3OneFormerSegmentationPyTorch
02 · Industrial · Camera networksShipped · 2026
Camera overlap analysis
Feature-matching pipeline using RANSAC to quantify field-of-view overlap between camera pairs. Produced per-pair overlap scores and camera-graph layouts that informed placement and eliminated blind spots.
Impact:Overlap graphs across venues · blind-spot elimination
RANSACFeature matchingOpenCVGraph layout
03 · Retail · Visual searchShipped · 2025
Visual product retrieval at marketplace scale
Built and deployed a state-of-the-art visual product retrieval model powering real-time product discovery from a phone photo. Surpassed Google Vertex AI multimodal embeddings on the same eval set.
Impact:+12.1% Recall@1 · +4.7% Recall@10 vs. Vertex AI multimodal
CLIPVertex AIPyTorchEmbeddingsRetrieval
04 · Logistics · VLMShipped · 2024
VLMs for parcel sorting & customs OCR
Explored Vision-Language Models to automate parcel-sorting validation in warehouses and extract structured text from customs labels. Replaced brittle template OCR with prompted VLM extraction.
Impact:Structured customs-label extraction · sorting validation in production
VLMLLaVAHugging FaceUnslothVLLM
05 · Aviation · OCRShipped · 2023
Vision Transformers for airport luggage
Built and optimized a Vision Transformer-based pipeline for video analysis of airport luggage footage, extracting flight numbers and license plates with high accuracy in operational conditions.
Impact:Flight # and plate extraction on live luggage video
ViTOCRTensorRTTriton
06 · Retail · Scan & GoShipped · 2022
Seamless basket recognition
Automated product recognition using feature-based matching of basket contents in retail. Layered consumer-behavior analysis on top of the recognition stream to surface buying patterns.
Impact:Basket-scan automation · behavior insights
Feature matchingPyTorchRetail
07 · Logistics · EdgeShipped · 2021
Address OCR + real-time barcodes
OCR and semantic analysis for sender/receiver address extraction. Optimized deep-learning pipelines for real-time barcode detection on small packages on a moving belt.
Impact:Real-time barcode detection · address extraction in production
OCRDetectionTensorRTEdge
08 · Healthcare · RetrievalShipped · 2019
CBIR for retinal fundus images
Siamese network trained on retinal fundus images to learn lower-dimensional representations minimizing intra-class and maximizing inter-class distance. One-shot retrieval at inference.
Impact:Published · BVM 2020 (Springer Vieweg)
Siamese networkOne-shotRetrievalTensorFlow
04/About
An applied AI engineering practice.
- 8 yrs
- in production ML
- 4
- industries served
- 1–2
- engagements at a time
- EU + US
- remote-friendly
WeConverge is an applied AI engineering practice focused on production computer vision and machine-learning systems. We work with operators in retail, logistics, industrial automation and healthcare — the places where models have to actually run, not just demo.
Eight years of shipping ML systems in production: state-of-the-art visual product retrieval at marketplace scale, real-time barcode detection on moving belts, depth + segmentation pipelines that auto-generate camera monitoring zones at venue scale, and VLMs for parcel sorting and customs OCR.
We take one or two engagements at a time — long enough to ship the system and hand it back working. No agency layer, no slide decks standing in for evidence. Just a small team that builds.
/ Founder
Azeem Bootwala — AI engineer, computer vision. Sr. CV Engineer at Basic-Fit, previously Data Scientist at Bol.com (Spot & Shop visual search) and six years of CV at Prime Vision. M.Sc. Medical Engineering, Erlangen-Nürnberg; published siamese-network retrieval at BVM 2020.
If you have a problem worth solving, we'd like to hear about it.
Contact · 05
Have a problem worth solving?
Tell us about it. We'll reply within one business day — usually with a clarifying question or two.
info@weconverge.nl
Delft · NL · UTC+1