Software engineering first · Colombo & Singapore
Inventory, POS, ERP and computer vision that already run real operations, extended with RAG, chat-with-your-data and private self-hosted AI. Your data never leaves your servers.
Four systems are live right now. Demo logins are printed on the product cards below, so you can judge the work rather than the pitch.
Track record
What we do
We ship working software, then add AI where it measurably pays for itself. Every engagement starts with the process costing you the most and ends with software running it.
The software your operation actually runs on, built to production standard and handed over with the source.
Your documents made answerable, with every claim anchored to the page it came from instead of confidently invented.
Cameras that count what matters, and models that run on hardware you already own.
Portfolio
Not slideware. Four systems are live with the credentials printed below. Open one, break something, and judge the engineering for yourself.
A graph-augmented RAG and document platform that compiles everything you upload into a maintained wiki, not just a chat box over a pile of PDFs. Documents are versioned, chunked and embedded, then compiled into concept pages, open questions and a knowledge changelog that compound as you add sources.
Built to run on your own hardware: air-gap compatible and tuned for open-weight 4B to 12B models, so sensitive documents never have to leave the building.
Retrieval pipeline
Camera-based detection and counting for vehicles, people and production-line items, with live annotated streams, unique-object counting and licence plate recognition. Works against CCTV and RTSP feeds, uploaded video and stills.
Every clip here is unedited VisionOps output, not a mock-up.
Multi-warehouse stock control with barcoding, transfers, reorder alerts and finance-grade reporting. Procurement through to AR/AP in one system.
demo login: admin@local · Passw0rd1
Fast point of sale that pairs with NueEdge Inventory. Sales, returns, shift management, receipts, pack-size selling and counter-sales shortcuts.
demo login: admin@local · Passw0rd1
Field-service and ERP workflows: job cards, technicians, parts, work orders, estimates, quality checks and full customer service history.
demo login: admin@iss.local · Admin123
A multi-agent research engine. Researcher, Analyst, Critic, Forecaster and Writer run in sequence to produce citation-backed reports with graded evidence and quantitative forecasts. Runs fully offline on local models.
A private, local-first engine that turns scattered notes, PDFs and decisions into a maintained, Obsidian-compatible Markdown knowledge graph. Ships as a CLI, a Python library, a sandboxed MCP server and a local web UI over a folder of plain files.
Knowledge engine
Most AI tools start from zero on every question. Ours do not. Both knowledge engines are built on the same principle: the second document about a topic should update what you know, not sit beside it as another unread file.
Security by architecture, not by policy.
Both engines default to open-source models running locally through Ollama, tuned for the 4B to 12B class that fits on hardware you already own. DocGraph is air-gap compatible: documents, embeddings and the compiled wiki stay in your own PostgreSQL and object storage.
The right model for each job, chosen for you.
Work is classified into fifteen distinct model purposes, from embeddings and reranking through to reasoning, vision and financial extraction, and each is routed to the model that handles it best. A 27B reasoning model is wasted on a classification call; a 4B model should not be writing your financial narrative.
Karpathy's LLM-wiki pattern, no Obsidian required.
The idea is usually demonstrated as a folder of Markdown edited by hand. DocGraph implements the same loop as a multi-user application: PostgreSQL, object storage and background jobs instead of a vault on one laptop. Upload a document and it is compiled into a page, reviewed, cross-referenced, and folded into concept pages that already exist.
The loop, every source runs it
Sources are preserved and versioned. Text, figures and images are extracted, chunked and embedded.
A wiki page is drafted, judged and committed, then merged into the concept pages it belongs to.
Ask questions against the corpus. Wiki pages give orientation; citations stay tied to source documents.
The knowledge base is checked for contradictions and drift. It reports, it never silently edits.
The same pattern, built twice for genuinely different situations: a shared platform for a team, and a private vault for one person.
A server application. Multi-user auth and role-based access, versioned documents, background job processing, and a web UI for chat, documents, wiki, authoring and admin. The knowledge base lives in PostgreSQL with pgvector, which is what makes permissions, versioning and concurrent access work properly.
A folder of plain Markdown you own outright, driven by a CLI, a Python library, and an MCP server that lets Claude Desktop or your IDE work against the vault directly. Open the same folder in Obsidian and the graph is simply there. Compatible with it, never dependent on it.
Safe by construction
Every path resolves inside the vault. Parent-directory traversal, absolute paths and symlink escapes are rejected. There is no delete tool and no shell tool. Writes are backed up and audit-logged, and read-only mode drops every write tool entirely.
Why NeuralsEdge
We come from shipping production business software in C#/.NET, Python, databases and deployments. AI gets added where it measurably helps, not as a demo trick that falls over on a Monday morning.
Every solution can run fully self-hosted: local models, local vector stores, on-premise vision processing. No customer data flows to a third-party API unless you deliberately choose it.
Four products are live for you to log into today, and the demo credentials are printed on this page. We would rather show working software than a pitch deck.
You get the source, the deployment scripts and the documentation. If we disappeared tomorrow, your systems keep running and someone else can maintain them.
Who you’d be working with
NeuralsEdge started in 2012 as a C# and .NET shop. That matters more than it sounds: most of what goes wrong with AI projects is not the model, it is the integration, the data and the operations around it, and all three are ordinary software engineering.
We work out of Colombo and Singapore, which means senior engineering at a rate that does not require a board sign-off, in a timezone that overlaps both Asia-Pacific and Europe.
Systems you can operate without us. Documented, observable, self-hosted, and handed over with the source.
Ship a demo and call it a deployment. If it cannot survive a Monday morning, it is not finished.
Enterprise software delivery, a decade before the word "AI" appeared on a single one of our invoices.
Pipelines, ETL and systems integration across finance, logistics and retail clients in four countries.
RAG systems, workflow automation and computer vision, built on the same engineering discipline as everything before it.
Seven products, four of them live for anyone to log into: inventory, POS, ERP, vision and two knowledge engines.
How an engagement runs
Most AI projects fail in the gap between a convincing demo and a system somebody has to run on a Monday. This is how we close that gap.
We sit with the people doing the work and document the process as it actually runs, not as the org chart says it does. You keep the map whether or not you hire us.
One workflow, end to end, against your real data on your own hardware. Small enough to throw away, real enough to prove the economics before you commit to the rest.
The full build: integrations, permissions, error handling, monitoring, and the unglamorous reliability work that decides whether this is still running in a year.
Source code, deployment scripts, documentation and dashboards, then a support window while your team takes the wheel. The goal is that you stop needing us.
Stack
Chosen per project, not per fashion cycle. Everything here is running in one of the products above, and if your team already has something that works, we build on it rather than replacing it.
Next step
Inventory, POS, ERP, CCTV, a shared drive nobody can search: describe what you have and we will tell you honestly what AI can add, what it would take, and roughly what it would cost. If the answer is that it is not worth doing, we will say so.