Claude & agentic systems
We design and ship reliable, production-grade agents on Claude — tool and MCP integration, context management, and architectures built to hold up well beyond the demo.
// applied-AI studio
Intelligent technology, applied.
We implement intelligent systems for real-world products — from production computer vision to agentic systems built on Claude.
01 — Services
What we do
LLM-driven development
We build software through large language models. For the past six months, every line we've shipped was produced through LLM-driven development — it's how we work, not a slide.
Applied computer vision & deep learning
Our foundation: hard, production machine learning. Real-time vision and deep-learning systems delivered under tight compute and reliability constraints.
02 — Selected work
Selected work
Client names withheld under NDA.
5×
Video compression
Cut stored video-stream size by 5× versus modern codecs, with no meaningful loss in quality.
All-weather
Parking occupancy detection
Vision-based parking-space occupancy detection that stays accurate in any weather — on compute-constrained edge hardware.
Real time
Gesture recognition
Human gesture recognition in real time on low-power hardware, fast enough for live interaction.
Super-human
3D medical segmentation
Segmentation of 3D medical scans at accuracy exceeding human performance.
03 — How we work
How we work
Two things make our implementations hold up. First, the discipline of shipping production machine learning — systems that must work on real data, on constrained hardware, every time. Second, we live agentic development daily: we architect and build through LLMs, so we know where they shine and where they break. We bring that combination to Claude implementations — focused on reliability, sound tool design, and architecture that survives contact with production.
Based On Tech is an applied-AI studio. We've spent years delivering production machine learning, and we bring that same engineering rigor to implementing Claude and agentic systems. We build for the real world — not for demos.