Ghassen Zaara® 000

Darmstadt, DE · CS @ TU Darmstadt

Ghassen ZaaraA man who doesn’t sleep untilhe solves the problem in question

Computer science student at TU Darmstadt. I like problems where the constraint is real. I love participating in hackathons and I am interested in every field computer science has to offer. Focused on software engineering, trying to improve my ML and cybersecurity skills.

Work

Selected work

01 / 05

ARIVA

Black Forest Hackathon · 2026 · team project

On-premise document processing for public administration.

Legal aid documents for Regierungspräsidium Freiburg, against a real requirements spec. No data could leave the premises under DSGVO, which ruled out every commercial API.

So the whole pipeline runs local. A vision-language model reads scanned PDFs and fills the legacy case forms. Pitched afterwards to Regierungspräsidium Offenburg as paid work.

  • Qwen3-VL
  • Ollama
  • FastAPI
  • Spring Boot
  • React 19
  • OpenCV
  • PostgreSQL

02 / 05

Viegtor

Futury Build Days · 2026 · 1st place

Strategic AI for patent and regulatory analysis.

Patents, regulations, tenders and competitor moves arrive as separate streams and get read by separate people. Viegtor distils all of it into one decision: build, invest, or ignore.

TODO · Viegtor: two or three sentences on what you actually built. Where did the data come from, what did the analysis do? Same length as the others.TODO · Viegtor tech stack

03 / 05

BrückenPilot

Interview project · 2026

German bridge inspection PDFs, turned into decisions.

Every German bridge inspection produces a 30-page PDF under DIN 1076. The data exists; nobody turns it into decisions.

BrückenPilot reads the PDF and returns a map. Condition scores, ranked damage with the inspector’s own photos, and the inspection trend projected forward. Extraction runs against Pydantic schemas at temperature 0.

  • FastAPI
  • React 19
  • TypeScript
  • Supabase
  • Gemini 2.0 Flash
  • MapLibre GL
  • PyMuPDF

04 / 05

VC AI Copilot

TUM.ai × Yellow Hackathon · 2026

One question, answered across four disconnected systems.

A VC firm’s knowledge sits in four places at once, CRM, meeting notes, Slack, email, and no system can answer what the current state of a company is, or what’s missing.

Affinity, Granola, Slack and Gmail collapse into one company object with source attribution and explicit gap detection. Every LLM call runs inside Supabase Edge Functions. No backend service.

  • Supabase
  • pgvector
  • Edge Functions
  • RAG
  • Affinity API
  • Gmail API

05 / 05

San Fo 薪火

EuroTech × HKTE Hackathon, Munich · 2026 · team project

Crowdfunding the Hong Kong shops that are quietly disappearing.

Hong Kong loses another family business every week. An ageing owner, no successor, shutters down for good. Banks will not lend without collateral and there is no legal route to raise equity from the public.

San Fo lets a neighbourhood fund the shop it does not want to lose: revenue-share agreements for individuals, tokenized SPVs for professional investors. KYC and KYB on both ends of every deal.

  • Swift
  • Rust
  • JavaScript
  • Tokenized SPV
  • KYC / KYB

Experience

Where I've worked

A blackboard covered in RISC-V cache addressing diagrams, with the tutor beside it.
Cache addressing, Übung 12

Tutor, Rechnerorganisation

TU Darmstadt, Fachbereich Informatik · Summer semester 2026

Weekly exercise sessions with 20+ students, in my fourth language, on RISC-V assembly and calling conventions, single-cycle and multicycle datapaths, ALU control decoding, pipeline hazards and flush penalties, cache addressing. I wrote the slide decks, ran the Testat scheduling and answered the course forum.

A lecture hall blackboard filled with NP-completeness reduction proofs.
NP-completeness, reduction proofs

Tutor, Algorithmen und Datenstrukturen

TU Darmstadt, Fachbereich Informatik · Summer semester 2026

Teaching 30+ students in my fourth language, marking 20+ assignments a week, and writing new coding exercises to push them past surface-level understanding.

Sorting and divide-and-conquer, radix sort, abstract data types, binary search trees, AVL and red-black trees, complexity and reductions. I redesigned one practical assignment on tree operations, black-height computation, join, AVL conversion and sorted merge, into a single unified 12-point exercise.

Coming soon

Research Assistant, Fraunhofer SIT / ATHENE

Starting October 2026

Studentische Hilfskraft in NLP research: authorship analysis, attribution and verification, style change detection, AI-generated text detection and LLM attribution. Classical feature engineering alongside embeddings, one-class and binary classification, ensemble methods.

Both modules ran in the same semester.

Awards

Hackathons

The winning team presenting Viegtor at Futury Build Days, one presenter on crutches.
Futury Build Days, Frankfurt

1st place

Futury Build Days, Viega Challenge

Frankfurt · 2026

Won the Viega challenge with Viegtor. Presented on crutches, broken ankle a week earlier.

Pitching Agnes to the jury at the TUM.ai Makeathon in Munich.
Pitching Agnes to the jury

2nd place

TUM.ai Makeathon, Spherecast Challenge

Munich · 2026 · 40+ teams, 500+ participants

Second place on the Spherecast challenge with Agnes: semantic supplier matching with a RAG pipeline and Pareto-front estimation across competing sourcing criteria. My first hackathon.

Community

Building the room

The HackathonHub Europe announcement card: a portrait against the European stars, captioned Ghassen Zaara, Ambassador for Frankfurt.
Ambassador for Frankfurt

Ambassador, HackathonHub Europe

Frankfurt

Regional ambassador for Frankfurt. HackathonHub connects student builders across Europe to hackathons and the teams running them.

TODO · HackathonHub: two or three sentences on what you actually do. Meetups? Recruiting teams? Point of contact for organisers? Right now this is a title with nothing behind it.TODO · HackathonHub: confirm scope (Frankfurt only, or Frankfurt and Darmstadt) and the start date

Toolkit

What I work with

Languages

  • Java
  • Python
  • C
  • TypeScript
  • JavaScript
  • SQL (PostgreSQL)
  • HTML/CSS
  • Bash

Frameworks

  • Spring Boot
  • React
  • Node.js
  • FastAPI
  • Express
  • Tailwind CSS
  • Pydantic

Testing & CI/CD

  • JUnit
  • Vitest
  • Playwright
  • pytest
  • Git
  • GitHub Actions
  • Jenkins
  • Maven

Cloud & tools

  • Docker
  • Kubernetes
  • Google Cloud Platform
  • AWS
  • Linux
  • PostgreSQL
  • Qdrant
  • REST APIs

Data & AI

  • pandas
  • NumPy
  • Matplotlib
  • OpenCV
  • RAG pipelines
  • vector search
  • embeddings
  • LLM APIs

Spoken

  • Arabic (native)
  • English, French, German (fluent)

Personal

Out Of Scope

It started with one PC and a timer. My brother and I took turns playing Minecraft, thirty minutes each, until I decided the real fix was a second machine. To convince my parents, I had to know exactly what to buy. So at ten I started reading about processors, memory speeds, GPUs and VRAM.

By thirteen I was building my own, sourcing components one at a time from wherever I could find them, aiming for a machine that could handle both gaming and photo editing. Photography had become the other thing I couldn’t put down. That PC still sits in a corner at home. I keep it there. It’s where programming started for me.

Mathematics, physics and the natural sciences pulled at me equally, which felt like an odd combination at the time. I liked running experiments and I liked solving problems on paper, and I never had to choose between them. I finished my baccalaureate at eighteen with 18.83/20, which converts to a 1.1 in the German system, ranked 27th out of 25,598 students nationally.

That left two paths: medical school, or computer science abroad. My family is full of doctors. I wanted to build something that was mine, so I took the second one.

I arrived in Germany with a B1 in German and some pocket money. I worked in restaurants and at events, studied for the telc C1 exam in two months, passed it on the first attempt, and applied to TU Darmstadt that same summer. Four months after landing, I was studying computer science in my fourth language. I set one rule for myself: pass everything the first time. The beginning was hard. I adapted.

Since then I’ve found hackathons, and I don’t just enter them, I farm them. New teams, new people, and problems I would never have run into on my own. Next comes NLP research at Fraunhofer SIT.

The thread has never changed. I want to know how things work, and then I want to build them.

Four piles. Open one.
  • A headland silhouetted against the sea at dusk.

Where I am from, and where I keep pointing the camera.

  • A brutalist concrete facade raking with afternoon light.

Architecture and coastlines, mostly. Rarely people, and never posed.

  • Boats moored in Portofino harbour below the painted houses.

Twenty-five places. Every caption here is where it was taken.

  • A pour-over coffee setup from above, grinder and scale in frame.

The one setup I fuss over every morning, and what comes out of it.