# Yassine Chettouch — Product Engineer

> Business software, built end to end.

I build web applications that save your team time and give your customers a better experience. Whether you’re launching a product or adding a major feature, I turn the business problem into software people can use and rely on.

Name: Yassine Chettouch
Also known as: yaschet
Role: Product Engineer
Based in: Rabat, Morocco
Canonical portfolio: https://www.yaschet.com

This document brings together the public portfolio, About page, working-together answers, contact details and complete published case studies. Project accounts and measurements are attributed to Yassine Chettouch; each study retains its demonstration context and sources.

## Contact and profiles

Source: https://www.yaschet.com/contact

- [Email hello@yaschet.com](mailto:hello@yaschet.com)
- [Book a 15-minute call](https://cal.com/yassinechettouch/15min)
- [WhatsApp @yaschet](https://wa.me/yaschet)
- [Contact form](https://www.yaschet.com/contact#form)
- [LinkedIn](https://linkedin.com/in/yassinechettouch)
- [GitHub](https://github.com/yaschet)
- [X](https://x.com/yaschett)
- [Resume PDF](https://www.yaschet.com/yassine-chettouch-resume.pdf)
- [Portrait of Yassine Chettouch](https://www.yaschet.com/images/avatar.png)

Tell me what you’re working on and where you need help.

### Working together

I build web applications and improve existing products. We can agree on a fixed scope or work together over time.

### Getting started

Share what you want to achieve, what’s already in place and any deadline. I’ll review the details, and we’ll agree on the scope, cost and timing before work begins.

## Selected work at a glance

Source: https://www.yaschet.com/case-studies

### [Verto: AI Translation Platform](https://www.yaschet.com/case-studies/verto)

A complete translation platform: turn scans into editable documents, review them beside the source and manage customers, credits and billing in one place.

Engagement: Independent product · SaaS
Role: Product Engineer
Year: 2026
Status: Released January 2026
Full text: https://www.yaschet.com/case-studies/verto.md

### [University Program Matching](https://www.yaschet.com/case-studies/clearway-phoenix)

University program matching that ranks thousands of options in under a second, with explanations and unmet requirements visible to the advisor.

Engagement: Client project · Internal tool
Role: Product Engineer
Year: 2025
Status: Delivered
Full text: https://www.yaschet.com/case-studies/clearway-phoenix.md

### [Student Applications & Document Collection](https://www.yaschet.com/case-studies/clearway-portal)

A student application portal that reads details from documents, saves progress across devices and delivers files and grades to the advisor's existing records.

Engagement: Client project · Client portal
Role: Product Engineer
Year: 2025
Status: Delivered
Full text: https://www.yaschet.com/case-studies/clearway-portal.md

### [Agency CRM & Billing](https://www.yaschet.com/case-studies/clearway-erp)

See each student's documents, responsible advisor and outstanding balance in one platform, shared by the application portal and program-matching system.

Engagement: Client project · CRM / ERP
Role: Product Engineer
Years: 2022–2025
Status: Delivered
Full text: https://www.yaschet.com/case-studies/clearway-erp.md

### There’s more to my work.

These case studies are a selection. Some client projects stay private, but I can talk through the problems I’ve solved, the decisions I made and [how I’d approach your project](https://www.yaschet.com/contact).

## About Yassine

Source: https://www.yaschet.com/about

I’m a Product Engineer based in Rabat, Morocco. I’ve been building web applications for businesses since 2022.

### I’ve always liked making things.

As a kid, I used the window ledge on our balcony as a workbench. I’d take motors out of toys and try to make little boats and planes out of ice-cream sticks.

In 2020, I started freelancing in graphic design while still at school. I went on to study software development in Rabat, followed by AI and data science.

### I’m particular about how things work.

I still notice the spacing, the proportions and the way a button responds. I’m just as interested in what happens after someone clicks it.

Someone should know whether their work has been saved. If a connection drops, they should be able to continue where they stopped. And when software makes a recommendation, they should be able to see why.

### I’ve learned to be clear from the start.

I learned early how much an unclear agreement can complicate a project. I now put the work, cost and review process in writing before development starts.

At handover, I give the client a walkthrough and instructions they can return to, so they can keep running what we’ve built and bring someone else in when they need to.

## Working together

Source: https://www.yaschet.com/#working-together

You stay involved in the decisions without having to manage every technical detail. We agree on the work and cost upfront, and you review working software along the way.

### What can I hire you for?

Complete web applications, new features and improvements to existing software. You can bring a clear technical brief or a problem you need help solving.

### Do you take on ongoing work?

Yes. I take on ongoing development and support, whether I built the original product or not. We agree on priorities, the time I can commit and the cost before we start.

### What do you need to get started?

Tell me what you need, why it matters and any deadline you're working toward. Share your existing software or brief if you have one. If the details aren't clear yet, we can work them out together.

### How do pricing and timing work?

I review the work before giving you a price and expected timeline. The written proposal explains what's included, the payment terms and what I need from you. I also explain running costs, such as hosting. We agree before I start.

### What is it like to work with you?

You work directly with me, see working versions and give feedback before release. I explain the choices and keep you updated on what's ready, what needs your input and what's next.

### What if we need to change the plan?

We talk about it before I build the change. I check what's already included, explain any effect on price or timing, and let you decide whether to proceed.

### What happens after launch?

I check the live product and hand over the code and instructions your team needs. We agree upfront on support after launch, including fixes for problems in my work. You can keep working with me or continue with another developer.

## Complete case studies and media

The full case studies follow, including the problem, contribution, delivery context, implementation decisions, results and demonstration notes. Image links retain their published descriptions and captions. Videos include titles, descriptions, duration where published, posters and playback links; these descriptions are not video transcripts.

---

# Verto: AI Translation Platform

> A complete translation platform: turn scans into editable documents, review them beside the source and manage customers, credits and billing in one place.

Canonical URL: https://www.yaschet.com/case-studies/verto
Date: 2026
Role: Product Engineer
Status: Released January 2026
Year: 2026
Domain: Document translation
Engagement: Independent product
Context: SaaS
Stack: Next.js, TypeScript, Supabase/PostgreSQL, Trigger.dev, Stripe, OpenAI, Tiptap, Google Docs API
Source code: https://github.com/yaschet/verto
Engineering handbook (PDF): https://www.yaschet.com/handbooks/verto-engineering-handbook.pdf

## The problem

A translation is not ready to deliver just because the words are in another language. Scanned reports contain tables, figures and names that must keep their meaning. Translators need to check the result against the original and deliver an editable document, while keeping track of the customer, the job and its cost.

## What I built

I designed Verto in Figma and built the complete translation platform. It translates scanned documents, checks the result against the source and gives translators a workspace to review and edit before delivery as Google Docs, DOCX or PDF. I also built customer records, agency branding, credits, Stripe billing and administration.

## Walkthrough and selected screens

> Video: Document Translation With Editable Tables: Walkthrough
> A photographed report becomes a Japanese translation, reviewed beside its source and delivered as a Google Doc with editable table cells.
> Duration: 1:40
> Poster: https://www.yaschet.com/images/verto/verto-film-poster.jpg
> Watch video (HLS): https://stream.mux.com/yiAwSuxIOVkCUZemf5jOzwfqwG01lH46A2GjtF8moZ8Q.m3u8

For a translation agency, the finished job is a document the customer can use. Translating a scanned financial report also means keeping figures in the right rows, preserving names and headings, checking the wording and producing a file that can still be edited. If the table falls apart, the translator still has a document to rebuild.

I designed and built Verto around that complete job. A translator uploads the source, receives a structured translation, checks it beside the original and edits the result before generating a native Google Doc or downloading DOCX and PDF. The platform also keeps the customer, job history, agency branding and credit transactions connected to that work.

I owned the product from Figma designs through implementation and release: the application, document model, AI pipeline, review editor, export generator, customer management, payments and administration. Verto is an independent product, released in 2026.

The engineering handbook documents the January 2026 handover.

The film follows a photographed English financial report into Japanese, then ends inside an editable table cell in Google Docs. It shows what the product delivers, from source document to a file the translator can continue working on.

## Preserve Structure Through Review

An automated translation still needs a translator's judgment. I put the source pages beside the translated document so the reviewer can compare a figure, check a name or correct a sentence without switching between disconnected files.

The editor supports changes to both wording and structure, including merging and splitting table cells. The reviewer is working on the document that will be delivered, so those corrections need to survive export.

I used a shared document model for paragraphs, headings and tables, including cells that span several rows or columns. The translation pipeline, Tiptap editor and export generator all work from that structure. Each correction has a defined place, and the generator has the information it needs to rebuild an editable table.

Names have an explicit preservation policy too. In the Japanese example, company and personal names stay in Latin script, making them directly comparable with the original. The product supports 68 target languages, including right-to-left output.

## Check Against The Original

The checking stage returns to the source page images after the first translation. It looks for issues such as missed content and table misalignment, then records its corrections and processing details. The original remains the reference throughout the job.

I built this as a separate pass because a completed AI response is not the same as a checked translation. The automated check helps find problems; the side-by-side editor gives the translator a place to make the final review before delivery.

## Carry Context Across A Long Document

A long report introduces a different problem. A term established near the beginning should not change meaning halfway through the document. Processing each group of pages in isolation would lose that continuity.

I carry names and terminology choices between batches, together with recent headings and a summary of the document so far. Later pages receive the context established by earlier ones. The film demonstrates that connection between pages 6 and 22 of the same document.

Arabic adds another requirement: reading direction must hold through the editor and the exported file, including the order of table cells. I handled direction in the document model and export logic so it remains part of the document throughout that journey.

## Generate A Native Google Doc

The difficult part of delivery was making the result editable outside Verto. I built a generator that turns the reviewed document model into native Google Docs paragraphs, headings, tables, headers and footers.

A table with merged cells cannot be recreated by placing its text in a sequence. A cell may occupy several rows or columns, and every later cell must land in the correct position. The generator first builds a complete grid in memory, marks the space each cell occupies, then creates and merges the cells through the Google Docs API. Right-to-left tables require their cell order to be handled explicitly as well.

This is the handoff the film closes on: the user clicks into a real Google Docs table cell, edits it and uses Tab to move between cells. The output is a working document. The same generation flow also supports DOCX and PDF, with the agency's logo, stamp and certification text.

## Build The Business Around The Document

I built customer management, translation history, account settings and administration alongside the document workspace. Each translation can stay connected to the agency's customer, while branding settings control the identity on the delivered file.

Processing also needs to work when the user leaves the page or a job fails. Translation and generation run through Trigger.dev background tasks, with saved stages, status updates and failure details. A user can return to the result for review and export instead of keeping the upload page open throughout processing.

Stripe purchases fund the credit balance used by those jobs. I implemented deductions, transaction history, credit adjustments and refunds for failed processing. Refund handling checks for an existing refund linked to the translation, so a retry can recognize one already recorded. Users can see the cost and refund in the same history as the job.

Together, these parts turn the document pipeline into a complete product for agency work: create a job for a customer, process the source, check and edit the translation, then deliver a branded, editable file.

*Verto is a personal project. The film uses synthetic agency records and existing completed translations; it demonstrates the workflow rather than live processing speed or customer adoption.*

Related work: [Student Applications & Document Collection](https://www.yaschet.com/case-studies/clearway-portal) applies document extraction to a guided customer workflow.

---

# University Program Matching

> University program matching that ranks thousands of options in under a second, with explanations and unmet requirements visible to the advisor.

Canonical URL: https://www.yaschet.com/case-studies/clearway-phoenix
Date: 2025
Role: Product Engineer
Status: Delivered
Year: 2025
Domain: Study-abroad advising
Engagement: Client project
Context: Internal tool
Stack: Laravel, PHP, Livewire, Alpine.js, MySQL, Redis, MeiliSearch, OpenAI

## The problem

Hundreds of students were waiting for university recommendations. Advisors spent hours cross-checking chaotic spreadsheets, then returned with options days later. Fees, admission rules and scholarships were buried in inconsistent notes. A missed condition could steer a student toward an unsuitable program or leave a better option unseen. These decisions could shape a student's education and finances.

## What I built

I designed and built Phoenix to rank thousands of programs in under a second, explain each match and flag unmet requirements. It connects a university management system, AI spreadsheet imports that can resume after a failure, and a workspace where advisors can compare options and save a student's shortlist during the call.

## Walkthrough and selected screens

> Video: University Program Matching: Walkthrough
> Follow the advising workflow with synthetic records: 19,874 programs ranked using 13 criteria, with explanations and saved student shortlists.
> Duration: 1:40
> Poster: https://www.yaschet.com/images/clearway-phoenix/phoenix-film-poster.jpg
> Watch video (HLS): https://stream.mux.com/002wHjSdVlFlojAAcI2q6HMRjSgcHejVR3LZ1y8KbWJo.m3u8

![The university program management table, with filters and expandable program details.](https://www.yaschet.com/images/clearway-phoenix/phase-1/01-programs-table.png)

Caption: Staff can find and update programs, fees and admission details in the application.

![Structured grade requirements for a university program.](https://www.yaschet.com/images/clearway-phoenix/phase-1/03-programs-modal-grade-requirements.png)

Caption: Each program's grade requirements give the matching system a way to check the student's academic fit.

![Scholarship types and categories maintained by agency staff.](https://www.yaschet.com/images/clearway-phoenix/phase-1/13-settings-scholarship-types-categories.png)

Caption: Shared scholarship types keep funding information consistent across programs.

Hundreds of students were waiting for university recommendations at Clearway, a study-abroad agency. Advisors spent hours piecing together options from spreadsheets where fees, admission rules and scholarship conditions were buried in inconsistent notes. Students finished a call and waited days for recommendations.

I built Phoenix so advisors could search and rank thousands of programs in real time, explain the recommendations and save a student's shortlist during the call. They can review grades, admission requirements, fees and scholarships together, then adjust the search as they learn more about the student.

The stakes were the student's education and finances. A missed admission requirement, misunderstood scholarship or overlooked program could lead them toward an unsuitable choice. I made the ranking transparent so an advisor could inspect the reasons, check the conditions and discuss the trade-offs with the student before recommending a program. The agency's reputation depended on that quality of advice.

The first matching tool relied on grade calculations and exact keywords. A search for economics could miss a business program because the wording was different. The database also lacked the detail needed to compare admission and scholarship conditions. Improving the search alone would not fix the information underneath it.

I rebuilt Phoenix as three connected parts. First, I designed the database and the screens staff would use to manage universities, programs, accommodation and scholarships. Then I built a resumable AI importer to turn the existing spreadsheets into connected records. Finally, I built the program search, scoring and advising workspace on top of that information.

I owned product design and engineering from requirements through production deployment inside the agency's existing Laravel platform: the database, management screens, AI imports, matching logic and complete advising interface.

## Rebuild The University And Admissions Database

The information was spread across university, program, accommodation and scholarship sheets. Names contained typos and abbreviations. Prices appeared as single amounts, ranges or several degree-specific fees in one cell. Admission and funding conditions were written as prose. Advisors had to interpret the information before they could even compare programs.

I designed the database to connect each program with its university, admission requirements, accommodation and available funding. The software could then distinguish what a scholarship pays for, who qualifies, how long it lasts and what renewal requires. Programs had defined grade and language requirements. Subjects belonged to broader fields, giving the search a way to connect related courses.

I also designed and built a management system for universities, programs, accommodation and scholarships. Staff could create, review, edit and remove records through the application. It included forms, searchable and filterable tables, expandable details, and settings for subjects, intakes, language tests and scholarships. Reusable requirements kept repeated information consistent, while AI autofill could prefill university and program details from a few inputs.

Staff could maintain the information themselves. A changed admission requirement or a new scholarship could be updated through the application and used in future recommendations.

## Import The Spreadsheets Without Starting Over After A Failure

The hardest part was interpreting what each cell meant. A row that looked simple in a spreadsheet could describe several records with different rules:

- A fee cell could list different charges for different degree levels, with inconsistent labels and formatting.
- Scholarship notes could describe several awards, each with its own coverage, duration and eligibility. Some entries ran together without spaces or line breaks.
- Accommodation notes mixed room types, facilities and extra charges.
- Language-test requirements could name a test without giving the minimum score.

Copying those cells into a database would preserve the confusion. I needed to separate the facts, put each one in the right field and retain the links between universities, programs, accommodation and funding.

I built an AI importer to process the spreadsheets row by row, interpret those entries, validate the extracted information and populate the new database. It created universities before their accommodation, programs and scholarships, preserving the links between them as spreadsheet references became connected records.

Running AI across thousands of rows made failures expensive. I saved completed transformations so unchanged rows could reuse their earlier results, and added checkpoints so an interrupted import could continue from saved progress. Temporary errors received a limited number of retries. Failed rows stayed available to inspect and correct.

That meant a failure near the end did not require starting the entire import again or paying for the same successful AI calls twice.

> Video: AI-assisted spreadsheet import
> An accelerated replay of a 33,614-row demonstration import, showing progress, saved work and failures that need attention.
> Duration: 0:20
> Poster: https://www.yaschet.com/images/clearway-phoenix/etl-film-poster.jpg
> Watch video (HLS): https://stream.mux.com/kJTw38g10201L1Ahz4jYM401rX00SFvJ01Se94llaIV3olPc.m3u8

![Import dashboard showing stages, failures and cost information.](https://www.yaschet.com/images/clearway-phoenix/phase-2/p2-archive-dashboard-hero.png)

Caption: The terminal dashboard shows which step is running, what has failed and how much processing has cost.

![Detailed progress and failure information for an import run.](https://www.yaschet.com/images/clearway-phoenix/phase-2/p2-03-editorial-signal-detail.png)

Caption: Saved progress and row-level errors make unfinished work visible before the import resumes.
*The 20-second film is an accelerated replay of a 33,614-row import using synthetic data.*

I built a terminal dashboard to monitor the run, including its current step, processing rate, errors and AI costs. It made a long-running import inspectable while it was happening.

The importer brought the agency's existing information into the application and could process future spreadsheets in the same format. The management screens handled day-to-day edits after import.

## Find Suitable Programs And Show Why They Match

With the information imported and maintainable, I could build the search and scoring system on top of it. A student interested in economics should be able to discover relevant business programs even when the course names use different words. I combined MeiliSearch with OpenAI embeddings so the search could match meaning as well as wording. I then built the rules that compare the results with the student's grades and preferences.

Finding a related subject was only the first step. A program could fit the student's interests while requiring grades they did not have, offering unsuitable funding or approaching its application deadline. I built a ranking model with 13 weighted criteria, including academic fit, subject relevance, scholarship potential, location and deadlines.

I made each match score explainable in the interface. Advisors can open the breakdown to see what raised or reduced the score, including penalties for unmet grade requirements. Critical conditions stay visible alongside the percentage, so the advisor has the information needed to judge the recommendation.

Each program is assessed against the same criteria. The breakdown makes the reasoning open to review: the advisor can see how the student's profile affects the ranking and which conditions still need attention. That gives them a consistent basis for comparing options, explaining the trade-offs and checking the recommendation before adding it to the student's shortlist.

![A machine-learning search returning related university programs.](https://www.yaschet.com/images/clearway-phoenix/phase-3/p3-01-semantic-search-machine-learning.png)

Caption: A search for machine learning returns related programs even when their names use different words.

![A program score broken down by matching criterion.](https://www.yaschet.com/images/clearway-phoenix/phase-3/p3-02-match-score-positive-breakdown.png)

Caption: Advisors can see which parts of the student's profile make a program a good match.

![Critical conditions shown alongside a program score.](https://www.yaschet.com/images/clearway-phoenix/phase-3/p3-04-critical-score-breakdown.png)

Caption: Admission conditions that need attention remain visible beside the match score.

![Programs assigned to a student with priority and status.](https://www.yaschet.com/images/clearway-phoenix/phase-3/p3-06-assigned-programs-workflow.png)

Caption: Selected programs stay on the student's record, with a priority and application status for each.
## Design The Workspace Around The Advising Call

An advisor needs to compare options while speaking with a student. I designed three connected areas: a searchable student list, the selected student's profile and program results, and filters with the student's assigned programs. Grades, study interests and recommendations can be reviewed together.

During a call, the advisor can adjust the student's information, narrow the results and inspect a program without leaving the workspace. Selecting a program adds it to the student's record. The advisor can then change its priority and update its application status as the student moves forward.

Recommendations had to arrive fast enough to keep the conversation moving. I optimized the search and results screen to return the first recommendations in under a second, fetching and rendering a small batch rather than every result. More programs load as the advisor scrolls.

I built the interface with Livewire and Alpine.js inside the existing Laravel platform. Switching students, changing filters and assigning programs update the relevant part of the screen without reloading the page. An advisor can keep refining the options with the student in front of them.

## Keep Recommendations Connected To Applications

I deployed the work in stages: the database and management screens, the spreadsheet importer, then the matching and advising interface. Each stage supported the next, from making the university information usable to putting recommendations in an advisor's hands.

The change is from hours of manual cross-checking and a follow-up days later to searching, comparing and shortlisting during the call. Advisors can respond to the student's questions as they arise, show why an option fits and surface conditions that need attention. Each shortlist stays attached to the student, ready for the team to manage the applications.

The team can also maintain the university information and import new spreadsheets without rebuilding the process. Fast recommendations depend on information staff can keep accurate and up to date.

The later student portal feeds extracted grades into the same student records. Document collection, program recommendations and application tracking can therefore use the same information inside the agency's platform.

*Clearway is an anonymized client identity. The films and screenshots use synthetic records and prepared replay data; displayed totals, timings and costs describe the demonstration.*

Related work: [Student Applications & Document Collection](https://www.yaschet.com/case-studies/clearway-portal) shows how documents and grades reach those same student records.

---

# Student Applications & Document Collection

> A student application portal that reads details from documents, saves progress across devices and delivers files and grades to the advisor's existing records.

Canonical URL: https://www.yaschet.com/case-studies/clearway-portal
Date: 2025
Role: Product Engineer
Status: Delivered
Year: 2025
Domain: Student onboarding
Engagement: Client project
Context: Client portal
Stack: Next.js, React, TypeScript, Laravel, PHP, AWS Textract, OpenAI, S3, next-intl, TanStack Query

## The problem

Student applications were scattered across WhatsApp conversations. Staff chased missing passports and transcripts, then retyped details already on the documents. Students needed to apply while still waiting for final results. The agency needed a complete application process that could handle documents arriving at different times and turn them into usable student records.

## What I built

I designed and built a four-language student portal that reads identity details and grades from uploaded documents, saves progress across devices and lets students apply before every file is ready. Students check the extracted details; advisors receive the files and grades in their existing records, without retyping them.

## Walkthrough and selected screens

> Video: Student portal walkthrough
> Follow an application from ID upload to the advisor's record: document prefill, cross-device resume, later uploads and a grade checked against its source.
> Duration: 1:40
> Poster: https://www.yaschet.com/images/clearway-portal/portal-film-poster.jpg
> Watch video (HLS): https://stream.mux.com/00k600fYHJNfFfMsU1vnVSlbNq5geNXDhXdZju29cSCRc.m3u8

![Student dashboard with saved application progress and document actions.](https://www.yaschet.com/images/clearway-portal/clearway-portal-poster-4k.png)

Caption: Returning students can continue their application and see which documents remain.

Students sent passports, IDs and transcripts to the study-abroad agency over WhatsApp. A passport arrived in one message, a transcript in another, and the remaining files came later. Staff had to piece each application together, chase what was missing and copy names, dates and grades into the agency's records. The information was already on the documents, but both sides still had more work to do.

The application also had to fit the student's circumstances. Someone waiting for final exam results still needed to start applying. Someone who began on a computer needed to continue on their phone. A single form that demanded every file in one sitting would not solve those problems.

I designed and built Portal to connect that entire journey. Students upload documents, check extracted details, save their progress and submit before every academic document is ready. They return to a dashboard to add the rest. Advisors receive the files and grades in the student records they already use.

I owned the product design and engineering from the client brief through production deployment: the four-language student experience, conditional application flow, document-processing integration and Laravel backend changes.

## Extract The Details, Then Let The Student Check

An ID already contains a name and date of birth. A transcript already contains grades. Asking the student to type them, then asking staff to enter them again, repeats the work and creates more opportunities for mistakes.

I connected the upload flow to document extraction using AWS Textract and OpenAI. An uploaded ID prefills identity fields; a transcript supplies grades for the student's academic record. Instead of treating the extracted text as unquestionable, the interface gives the student a chance to check it. Identity fields remain editable before confirmation.

Source badges make that review practical. A name filled from an ID is marked as coming from that ID, so the student knows why it appeared and what to compare it with.

The same connection continues into the advisor's workspace. Extracted grades stay linked to the original transcript, so an advisor can check the numbers against the document. That matters beyond filling in a profile: Phoenix, the program-matching system I built for the same agency, uses those academic records when evaluating university options.

The film follows one prepared example all the way through: Mathématiques **17,50 / 20** on the transcript becomes **17.50 / 20** on the advisor's record, with the source still available to inspect.

![Identity details prefilled from an uploaded ID, with source badges.](https://www.yaschet.com/images/clearway-portal/clearway-portal-id-prefill-4k.png)

Caption: The uploaded ID fills the name. The source badge explains where it came from, and the student can correct it before continuing.
## Save Progress Beyond The Device

The student sees a focused question or upload at each step. Behind that simple interface, the next step depends on their answers: the degree they want, the qualifications they already hold and whether they have a passport.

I defined the steps and their conditions in typed configuration, separate from the screen components. The flow can ask for the relevant academic documents without sending every applicant through the same questions. Moving back, changing an answer and returning later all have to follow those conditions.

Progress is saved to the backend, including the answers and current step. Returning students recover the application from their account rather than relying on one browser's local storage. The film shows an academic goal saved on desktop, then restored in a separate session on an iPhone.

I built the experience in English, French, Arabic and Darija, including right-to-left layouts. The guided flow, document instructions and review screens follow the student's chosen language.

![A saved academic goal resumed on an iPhone.](https://www.yaschet.com/images/clearway-portal/clearway-portal-phone-resume-4k.png)

Caption: A goal saved on desktop is restored in a separate iPhone session.
## Let Students Apply Before Every Document Is Ready

An application can be ready to begin before its documents are complete. Students finishing secondary school may still be waiting for final results. Requiring every transcript at the start would block them for something they cannot yet provide.

I separated submitting the initial application from completing the document collection. Students provide their core identity information and submit, then use the dashboard to see what has arrived and add the remaining files as they become available. Finishing the guided application does not make outstanding documents disappear.

The processing states matter here too. A successful upload means the file arrived, not that its contents have been read. I made uploading, processing, completion and failure visible as separate states. Documents that need extraction follow that path; administrative files that only need storage do not pretend to have been processed.

![Student dashboard showing submitted and outstanding documents.](https://www.yaschet.com/images/clearway-portal/clearway-portal-document-status-4k.png)

Caption: A submitted transcript stays visible alongside documents the student still needs to provide.
## Build On The Records The Agency Already Uses

A polished student portal would still leave the job unfinished if staff had to copy its submissions into another system. I connected Portal directly to the agency's existing customer records, including the attachments and academic grades that advisors need for their work.

My earlier work on Phoenix had exposed the difficulty of building this kind of continuous, interactive journey with our Livewire setup. I chose Next.js for the student interface while keeping Laravel responsible for the established records, validation, authorization and document processing.

That meant adapting the new application to the existing data model. Saving a step preserves the working application state and updates the relevant customer fields. Documents and extracted grades attach to that same customer. There is no second student database for the agency to reconcile.

Passports and transcripts also needed deliberate access controls. Authenticated requests run through Next.js server actions, with session tokens held in HttpOnly cookies. Laravel checks that a requested attachment belongs to the signed-in customer before issuing a temporary S3 download URL.

![The extracted grade 17.50 beside its original transcript in the advisor ERP.](https://www.yaschet.com/images/clearway-portal/clearway-portal-transcript-erp-4k.png)

Caption: Mathématiques 17,50 on the source transcript becomes 17.50 in the advisor's record. Both remain available together, so the grade can be checked without typing it again.
The main maintenance lesson is to keep navigation, required documents and completion tied to the same rules. Some rules remain duplicated; consolidating them would make future changes to an education path easier to carry through consistently.

## From Application To Advisor

The result is a connected application process: students can start with the documents they have, check prefilled details, continue on another device and supply missing files later. Advisors can open the resulting record with the documents and grades already attached. The handoff no longer depends on staff re-entering what the student has already provided.

The walkthrough follows that handoff from the first ID upload through the matching grade in the advisor's record.

*Clearway is an anonymized client identity. The demonstration uses synthetic student data and prepared document responses.*

Related work: [University Program Matching](https://www.yaschet.com/case-studies/clearway-phoenix) uses these academic records to explain suitable programs.

---

# Agency CRM & Billing

> See each student's documents, responsible advisor and outstanding balance in one platform, shared by the application portal and program-matching system.

Canonical URL: https://www.yaschet.com/case-studies/clearway-erp
Date: 2026-10-30
Role: Product Engineer
Status: Delivered
Years: 2022–2025
Domain: Study-abroad operations
Engagement: Client project
Context: CRM / ERP
Stack: Cash Plus, Laravel, PHP, Livewire, Alpine.js, MySQL, Redis, MeiliSearch

## The problem

An application does not end when a student submits a form. Staff still have to collect documents, coordinate advisors, invoice services and track payments. Those records need to agree: a payment must change the balance, a case must have a responsible advisor, and new student-facing tools must work with the records already in use.

## What I built

I designed and built the agency's CRM and billing platform, from customer records and staff permissions to invoices, payments and receipts. I made balances and payment statuses follow the underlying entries, then integrated the student portal and program-matching system around the same customer record.

I built the CRM and billing platform that lets a study-abroad agency follow each student's application and payments in one place. Staff can see missing documents, the responsible advisor and the outstanding balance without piecing together separate records.

The student's record connects their documents, grades, selected programs, invoices and payment history. Balances are calculated from the services, discounts and payments recorded, so staff can check how an amount was reached.

I built the initial platform, then returned for several rounds of client-requested improvements between 2022 and 2025. As the sole engineer, I led discovery, product design, implementation and deployment. The work included the data model, staff interfaces, billing rules, permissions and integrations with the agency's later student portal and university-matching system.

## Keep One Student Record Across The Workflow

The student is the common reference across the agency's work. I connected identity details, uploaded documents, grades, the assigned advisor and selected programs to that customer record. Invoices and payments belong to the same customer too.

That relationship became important as the platform grew. The student portal writes application details, files and extracted grades into the existing record. Phoenix reads the academic information for program matching and saves selected programs back to that student. The student-facing and staff-facing tools contribute to the same case.

I treated compatibility with the existing agency workflow as part of each new product. A new interface had to respect the fields, staff assignments and access rules already in use. It could improve how a student applies or how an advisor searches without requiring a separate copy of the student.

## Derive Invoice Balances From Payments

The difficult part of billing is keeping the answer correct as the record changes. A student can pay part of an invoice, receive a discount or have an invoice cancelled. A manually maintained status can drift away from the amounts staff are looking at.

I connected each invoice to its service items, discount and individual payment entries. The system calculates the invoice total after discount, adds the recorded payments and derives the remaining balance. Partial payment stays visible as partial payment. A cancelled invoice keeps its cancellation state; unpaid invoices are checked against their due dates.

The payment history remains part of the case. Staff can follow an entry to its invoice and customer, and receipts use the related invoice, customer and company details. This gives the displayed amount an explanation rather than leaving it as an isolated number.

I also built expense records and the surrounding staff workflows. Customer operations and billing live in the same application, with their own rules and relationships.

## Keep Access Tied To Responsibility

Sharing a platform does not mean every staff member should see every case. Consultants work with their assigned customers, while managers have views of the records they manage.

I applied those assignments to customer queries as well as the interface. The application retrieves the records appropriate to the staff member's role and responsibility, instead of loading everything and relying on a hidden button or filtered screen.

Staff also have management screens for universities, programs, scholarships and accommodations. The agency can maintain the information used for recommendations inside the same operations platform that holds the student's case.

## Extend The Platform Without Splitting The Records

The work took place in separate engagements. As needs changed, I added capabilities around the existing customer and financial relationships rather than treating each request as a separate application.

Phoenix adds program search, ranking and shortlists. The student portal adds guided applications, document uploads and extraction. Each is a distinct product experience, with this CRM providing the shared student records and staff access rules behind it.

The result is a connected path from a student's submitted documents to the advisor's program selections and the agency's invoices and payments. Staff can follow the application and its financial history without recreating the student for each tool.

*Clearway is an anonymized client identity. The ERP, student portal and program-matching system were built for the same agency.*

Related work: [University Program Matching](https://www.yaschet.com/case-studies/clearway-phoenix) shows the program catalog and shortlist workflow built on this foundation.

---

## Contact Yassine

[Email hello@yaschet.com](mailto:hello@yaschet.com) · [Book a 15-minute call](https://cal.com/yassinechettouch/15min) · [Contact page](https://www.yaschet.com/contact)