WebCorrect.pro
A software development firm that takes projects from research through to the app store and the production server — native mobile apps, applied machine learning, search platforms, and the infrastructure underneath them.
Visit webcorrect.proWhat they build
Across mobile, machine learning, and backend systems.
Mobile applications
Native Android and iOS apps in Kotlin and Swift, taken through signed builds, staged rollouts, and both app stores — not prototypes that never leave a laptop.
Applied machine learning
Speech recognition, computer vision, and language models trained on GPU clusters, then quantized and exported to run entirely on a phone.
Web platforms & APIs
Python API services on PostgreSQL, deployed behind nginx with TLS on cloud infrastructure they provision and keep running.
Selected work
Three recent projects, described by what they required.
Speech recognition
A speech model for a language nobody had trained one for
Fine-tuned a transformer speech model for a historic liturgical language with almost no training data — which meant building the corpus first, then running a disciplined evaluation ledger against word-error-rate, negative results included.
Stack: PyTorch, Transformers, cloud GPU training, ONNX, on-device inference, Kotlin, Swift, CoreML.
Search & data
Semantic search across a multilingual historical corpus
A live search engine over ancient-language texts in Greek, Latin, and Syriac, combining full-text and vector similarity so queries match meaning rather than only words — fed by an ingestion pipeline for messy OCR sources.
Stack: Python, FastAPI, PostgreSQL, pgvector, sentence embeddings, spaCy, ETL, nginx + TLS.
Computer vision
Real-time vision and a language model, both on the phone
A consumer app running instance segmentation on camera input with narration from a quantized on-device language model. When the standard export toolchain produced silently wrong output, they diagnosed it and re-implemented the decode and suppression stages in the app runtime.
Stack: instance segmentation, ONNX Runtime, quantization, on-device LLM, Android, iOS.
How they work
- Understand the real problem. Scope honestly at the start rather than discovering the constraint in month three.
- Prove the hard part first. Build the piece that might not work, and measure it honestly — including when the answer is no.
- Build it properly. Tested code, reproducible builds, and documented runbooks, because the handover is part of the job.
- Ship and keep it running. Store submissions, deployments, certificates, and the unglamorous maintenance afterwards.
See the full site
The full capability list, selected work, and how to start a project all live on the WebCorrect.pro site.
Go to webcorrect.pro