Software systems
Applications, APIs and foundations designed as one coherent system — from interface to production.
Independent engineering lab Small by design.
Small by design.
Technical
by nature.
LeLab is where software, AI systems, infrastructure and developer tools are explored, built and tested against real constraints. One engineering view, from first decision to operation.
◇Whole-system view ⌘Built from real friction ⊞Public proof of work
AI + data systems
Agents, retrieval and integrations sized to the actual problem, with explicit boundaries and controllable data.
Infrastructure + tooling
Deployment, observability and delivery tooling that stays understandable, inspectable and under control.
01 / From the Lab Real friction.
Real friction.
Systems from the Lab.
Each character represents a real tool. The illustrations share a world; the projects solve concrete problems in software, infrastructure and AI.
Lightweight monitoring
Monitoring without the operational weight.
A compact monitoring system for services, incidents and public status — built because the monitoring layer should not become another platform to operate.
SvelteKit app foundations
Stop rebuilding the same foundations.
A reusable SvelteKit foundation that turns recurring production decisions — auth, structure, quality and delivery — into a solid starting point.
AI coding orchestration
Agents need a system around them.
An orchestration workflow that separates implementation from review, ties work to issues and keeps the final merge decision with a human.
Lightweight deployment control
Deploy without adding another platform.
A lightweight deployment layer built around existing Compose services: explicit actions, fewer moving parts and an inspectable path to production.
Personal retrieval system
Retrieval without giving up the source of truth.
A lightweight document memory for AI assistants: local files stay authoritative while a rebuildable index makes their contents retrievable through MCP.
Scroll-driven motion
Motion without the animation stack.
Scroll-triggered motion built on browser primitives, CSS transitions and a small Svelte 5 integration — no framework-sized dependency for a narrow job.
Vaka
Monitoring without the operational weight.
01 / THE FRICTION
Watch services without adding a large platform to operate.
02 / THE DECISION
Keep monitoring focused on availability, latency and incident history. A Rust service provides the foundation for a compact deployment.
03 / THE SYSTEM
In progressA compact monitoring system for services, incidents and public status — built because the monitoring layer should not become another platform to operate.
02 / Interactive experiment
Don't take the model's word for it.
A focused decision engine, not another chat interface. Try a support ticket, inspect the result, then change the questions and see what the system actually does.
EXPERIMENT / 001 JEV-COMPATIBLE · BROWSER PREVIEW + SEPARATE PRIVATE LAYA ACTION
Start with a scenario, or write your own ticket. Decision rules remain locked until you enter edit mode.
107 / 4000 characters / 4000 FR / EN supported
Browser simulation · no AI inference Live inference sends this ticket to the private model; use test data only.
Waiting for a ticket.
Run the simulation to see classification, priority and routing.
Decisions are proposals, not autonomous actions. HUMAN IN THE LOOP / 01
CONFIGURATION / ADVANCED
Change the decision, not the model.
Customize the questions sent to Laya, the available categories and the business-rule thresholds. Nothing is sent until you explicitly choose private Laya inference.
one key: description per line
one level per line
BUSINESS RULES / OUTSIDE MODEL
Critical tickets should always remain subject to explicit safety and escalation policy.
PAYLOAD / JEV FORMAT
The browser simulation never makes an API request. The separate Laya action sends state and typed questions through the server-side private proxy; application thresholds stay outside the model.
Why this exists. An LLM isn't the answer to every decision. A narrow model plus explicit rules can be easier to understand, operate and audit.
NO MAGIC BLACK BOXES / LELAB03 / The engineering view Explore widely.
Explore widely.
Build with intent.
The Lab tests ideas against real constraints. New technology is welcome; complexity still has to earn its place.
A real problem
comes first.
A service to watch. Foundations rebuilt too often. Agents that need structure. Friction gives the work its direction.
Use less when
less is enough.
Choose the model, platform and architecture for the job. Every dependency and abstraction adds something to understand, secure and operate.
Production is part
of engineering.
Test it. Deploy it. Observe it. Maintain it. What ships should be understood well enough to operate when conditions change.
04 / Applied engineering The same thinking.
The same thinking.
Real-world constraints.
The same principles have to survive existing software, business needs, users and operational limits. Client work is where they meet the real world.
ANRISOL / AI SYSTEMS A smaller decision layer.
A smaller decision layer.
A better-sized system.
Reshaping an existing AI workflow so each decision uses the smallest layer that fits.
01 / CONSTRAINT
A larger language model was handling decisions that did not need its cost or complexity.
02 / ENGINEERING DECISION
Introduce Jev for focused decisions while preserving the broader workflow and its integrations.
03 / RESULTING DESIGN
Keep larger-model calls for the steps that benefit from them. Reduce unnecessary model dependence without fragmenting the system.
05 / Behind the Lab 
Ludo Software engineer · Independent builder
LeLab is the independent engineering lab of Ludo. It brings together public projects, technical experiments and selected real-world work across software, AI systems, infrastructure and developer tooling.
These projects are not a technology showcase. They are where recurring friction becomes a system, new ideas are tested and engineering decisions are made visible.
Open to selected engineering roles, collaborations and independent work.
More about Ludo06 / Ask the Lab Curious about
Curious about
how it works?
A small technical FAQ about the projects and decisions behind LeLab. With your consent, record a question for private Whistle transcription. LeLab does not store the audio; review the transcript before choosing to send it to OpenAI.
01 / Text 02 / Cactus Whistle 03 / GPT + grounded context
ASK / PUBLIC CORPUS
Ask a free-form question for a GPT answer grounded in selected public sources. Questions are sent to OpenAI via the server; LeLab does not keep conversation history. Do not include sensitive information.
Submitted questions go to OpenAI through the server · LeLab keeps no conversation history. Voice audio goes to private Whistle only after you press record; review the transcript before sending it. Do not include sensitive information.
Ready · GPT answers from the selected public corpus
Microphone permission is requested only after you press record. BUILT TO BE INSPECTABLEContact
hello@lelab.devHave something interesting in mind?
LeLab is open to selected engineering work, collaborations and technical opportunities — especially where the problem is worth understanding before choosing the solution.