01
Private AI and knowledge systems
Assistants that answer from your own documents, inside a boundary you have agreed.
See deliverables and the pilot
AI engineering · Powered by MediaElegant
AI systems built for the way your business works.
We build private AI, automate business workflows and develop custom software. Read plain explanations of how AI works and explore the systems we have helped build.
Illustrative workflow · authored sample
A rough client brief becomes a content plan a strategist can review.
1 · Request
What the system is asked to do, and by whom. Nothing starts without a request a person can read.
2 · Sources
The documents the system is allowed to read for this task. It works from these, not from memory.
3 · Draft
A structured first version. Each line points to a source, or is marked as an assumption to confirm.
4 · Review
A configured step where a person decides. This is a sample policy: each real workflow sets its own.
5 · Result
The output in an editable format, with its sources attached. Nothing leaves without the review above.
Sample task: Fictional bakery chain wants a four-week spring content plan. Instagram and a newsletter, small budget, launch in April. Here is their brief and last year's recap. Interactive controls load with the page. Nothing here is a live system.
Illustrative workflow with authored sample data. No live model runs and nothing is sent anywhere.
Solutions
Four ways we help.
Each starts with a conversation and a measurable pilot. Scope and price are written down before work begins.
02
Business automation and integrations
Connect the tools your team already uses so repeated work moves without re-keying.
See deliverables and the pilot
03
Custom software and AI products
Web, mobile and internal tools with APIs, built with a spec, tests and a clear scope.
See deliverables and the pilot
04
Model evaluation, adaptation and deployment
Measure a model on your task first, improve it the cheapest way that works, then serve it.
See deliverables and the pilot
Selected work
A product people can use, and how it is built.
Each project story shows the problem, our role and the real status. Open the system below to see what each part does and why.
Live · built by MediaElegant
Multilingual AI travel guide
A traveller describes the trip they want in ordinary language and gets a day-by-day draft built from a searchable directory of places, with every price labelled as an estimate.
Example itinerary
6-day draft, first 4 days shown
~8,200 MAD est.
- 1Marrakech: medina and souksWalking tour0 MAD
- 2Aït Benhaddou and the AtlasShared driver~900 MAD
- 3Sahara: camel and campDesert package~1,400 MAD
- 4Fès: the old medinaTrain, 2nd class~120 MAD
Inspect the system
How Multilingual AI travel guide is put together
- What it is
- Open data and checked sourcesThe raw place data the directory starts from.
- What flows through it
- Public data comes in. Each record keeps where it came from and when it was last checked.
- The decision
- Every listing shows its source and last-checked date, so a reader can judge it.
Read the whole system as a list
- Open data and checked sources. The raw place data the directory starts from. Flow: Public data comes in. Each record keeps where it came from and when it was last checked. Decision: Every listing shows its source and last-checked date, so a reader can judge it.
- Data pipeline. Loads the open data into PostgreSQL with PostGIS for location queries. Flow: Raw records go in. Deduplicated, validated records come out. Decision: Nothing reaches the planner until it has been deduplicated and validated.
- Hybrid retrieval. Keyword and vector (pgvector) search over the directory. Flow: A traveller's request goes in. Candidate places from the directory come out. Decision: Combine keyword and vector search, so exact names and loose descriptions both find places.
- Day-by-day planner. Turns a natural-language request into a day-by-day draft. Flow: Days, budget and retrieved places go in. A draft grouped by area, with travel time allowed for, comes out. Decision: AI proposes, sources decide: the planner works from the directory, not from memory.
- Price estimates. The prices shown next to a plan. Flow: Estimated costs are attached to plan items and shown to the traveller. Decision: Every price is labelled as an estimate, never a quote.
Blog
Practical AI, explained plainly.
Standalone articles on what AI is, how it works and how to use it in a business. Each one says what it is based on.
Guide · beginner · · 6 min
How to choose your first business AI workflow
Choose one useful AI pilot by defining the task, checking the inputs, setting a review boundary and measuring the result.
- business-ai
- automation
- getting-started
Explainer · beginner · · 4 min
How to give AI a useful brief: goals, context and clear constraints
A vague request leaves the important choices to the model. A good brief states the goal, the audience, the facts and the limits, and asks for what is missing.
- ai-basics
- prompting
- getting-started
Explainer · beginner · · 5 min
How language models work, and how they differ from AI applications
A language model predicts text. The chat app, search tools and business integrations around it are separate things. Knowing the difference explains a lot of confusing behaviour.
- ai-basics
- language-models
- getting-started
How we work
Engineers decide. Agents do the volume. Tests are the gate.
AI makes code cheap to write. What stays expensive is knowing what to build and what is safe to ship, so an engineer owns both ends.
- 01Engineer
We write the spec
Architecture, API contracts, data flows and acceptance criteria come from an engineer before any agent runs.
Output: spec, API contracts, acceptance criteria
- 02AI
Agents do the volume
Scaffolding, refactors, test generation and documentation go to AI agents, in parallel and fast.
Output: scaffolding, refactors, tests, docs
- 03Both
Tests are the gate
Unit, API, end-to-end and manual checks decide what ships. Nothing merges on a hunch.
Output: test results the team can read
- 04Engineer
We review every diff
Security, access control, cost, failure modes and edge cases: the judgement that years of production teach.
Output: reviewed diffs, with changes requested where needed
Where it runs is a choice.
We write down where every stage touches your data before we build. Read the security approach.
A server you own
Runs in an environment you control.
A dedicated environment
Set up for you alone, with responsibilities written down.
An approved API provider
A hosted model under terms you have reviewed.
Powered by MediaElegant
Backed by a studio that has shipped 40+ products.
Nupgrade is MediaElegant's AI engineering company. Design, mobile, cloud, SEO, QA and maintenance are available from the wider team when a project needs them. Figures are as MediaElegant publishes them.
- 8+
- years shipping
- 40+
- products launched
- GMT+1
- overlaps Europe and the Gulf
- Remote-first
- rooted in Marrakech
- Software and SaaS engineering
- Web development
- Mobile apps
- Desktop applications
- Cloud and DevOps
- Developer tools and automation
- Accessibility engineering
- Content protection
- QA and testing
- SEO and digital marketing
- API development and integrations
- Maintenance and support
Questions
Things people ask before they write.
What does it cost?
We scope and price in writing before work starts. We do not publish a price list, because the right answer depends on your process, your data and how much has to be integrated. A paid scoped assessment is available when the question needs real investigation first.
Will AI make mistakes with our data?
Models can be wrong, so we design for it. Where it matters, answers point to their sources, risky actions wait for a person to approve them, and the pilot has an acceptance test written before it starts. We do not promise zero errors.
Can it run privately, without sending our data to a public chatbot?
Often, yes. The options are a server you own, a dedicated environment, or an approved API provider under agreed terms. What is private depends on every stage: extraction, embeddings, retrieval, model inference, logs and backups. The security page explains how we scope that boundary.
Who does the work?
Engineers working inside MediaElegant's remote-first team in Marrakech. An engineer writes the spec and reviews every change. AI coding agents help with volume work under that review. We name the people on a project once they and the client agree.
Which tools do you work with?
The tools your team already uses: email, chat, CRMs, spreadsheets, document stores and helpdesk software, usually through their APIs. We check what a given integration supports during scoping rather than assuming it.
What do we own at the end?
That is agreed in writing before the work starts. Our aim is that you receive the code, workflows, documentation and a runbook for what we build for you.
What time zone and languages?
GMT+1, which overlaps Europe and the Gulf. We work in English and French. If you need Arabic or Darija, we test the specific task with your real material before promising it.
One question
Which manual process currently takes the most time from your team?
Answer in a sentence or two. An engineer replies with where we would start, what it takes and what we would not automate.