Software, servers, hardware integration, industrial communication and applied AI.
Twenty-five years' experience, beginning in high-performance computing. I work on projects that sit between technologies rather than inside one.
Connecting industrial equipment and sensors to servers. Building applications and APIs. Integrating hardware with software. Standing up self-hosted systems. Adding AI capabilities to software that already exists.
I look at a problem as a complete system rather than as one piece of software to be written. How the hardware, the network, the server, the security model and the deployment behave together — in a real environment rather than a demonstration one — is usually where a project succeeds or fails.
Desktop, mobile and web in Flutter, including packages published for other developers.
Python service architectures, APIs, job and worker systems, Linux servers, Docker.
Machine signals, gateways, microcontrollers and edge devices.
Image embeddings and similarity search, computer vision, face detection and recognition, edge-AI accelerators such as Hailo and Coral, and LLM and agent integration including MCP.
I am drawn to problems that are practical and technically awkward — the ones with no off-the-shelf answer. I prefer to understand the whole requirement first, then design something reliable, maintainable and practical to deploy.
I am equally comfortable working within an existing system or building from scratch, and I run machine-learning inference on modest on-premises hardware by preference rather than depending on hosted services.
A self-hosted media and machine-learning platform, and the largest thing I have built. Twelve repositories under one workspace: three independent FastAPI services for authentication, storage and computation; a plugin-based worker system for the processing and inference work; MQTT job orchestration with a queue and worker coordination; versioned storage with a full audit trail; and JWT authentication.
On the client side, a Flutter application for mobile and desktop, Python and Dart SDKs, and command-line tools in both languages — all against the same services. It runs on a spare machine or a Raspberry Pi, and nothing it holds leaves the network it runs on. Releasing shortly under Cloud on LAN Apps.
A master–worker media processing and machine-learning toolkit. Nine plugins covering image and video processing, CLIP and DINOv2 image embeddings, face detection and recognition, and HLS streaming — with FastAPI integration, an async priority job queue with MQTT, and hardware acceleration for the Raspberry Pi 5 and Hailo-8. MIT licensed.
My own products: applications that run on a server you control, on your own network, with nothing sent to a third party.
Six Flutter and Dart packages published under the Cloud on LAN Apps organisation, a verified publisher on pub.dev. All MIT.
Systems and methods of
buffering and accessing input data for convolution computations
A buffering scheme that reduces the convolution computations of a neural-network layer to a single matrix multiplication, with no data movement in memory — arranging the input data so the filters collapse into linear arrays and the buffer refreshes by advancing read pointers rather than copying. Filed with five co-inventors.
LinkedIn · GitHub · GitLab · Patents · anandas.cc.cc