Abdelkerim Dassi

AI Engineer: LLM systems, RAG, agents, edge computer vision
Tunis, Tunisia · open to relocation: Gulf, EU, UK, US abdelkerimdassi@gmail.com +216 55 683 474 abdelkerimdassi.com linkedin.com/in/abdelkerim-dassi github.com/Abdelkerim-Dassi
Notice period: 21 days Work authorization: visa sponsorship required Arabic (native) · English (C1) · French (C1) · German (basic) · teach in all three

Profile

AI engineer who ships to production: RAG, LLM agents, agent evaluation and edge computer vision. In AI since 2021: internships in Zurich and with NTT DATA Munich during my engineering degree, then full-time from 2025. Today I am AI Team Lead at 1morething Studio, owning the AI roadmap across its product portfolio and building hands-on: edge vision live in a Riyadh coffee chain, an Arabic RAG assistant for FAO in Egypt that has answered 15K+ questions, and an on-premise legal-document pipeline for a ministry. In summer 2026 I built and calibrated RL agent-evaluation tasks for Anthropic's frontier models at Originator. Looking for a full-time AI / ML engineering role abroad.

Experience

AI Team Lead
1morething Studio · Tunis, Tunisia · Hybrid
Feb 2026 – Present
  • Own the AI roadmap for every product in the venture portfolio: validate architecture and model decisions, set the AI service standards, and review the AI services the team ships. Mentor 2–4 engineers.
  • Absar, edge vision (coffee chain, Riyadh): built all 11 on-device SOP checks (hygiene gear, food safety, drive-thru) live on NVIDIA Jetson, 8 cameras at 16 FPS. A GPU benchmark I ran gave 17× faster inference (3.9 s → 225 ms) and set the box's architecture.
  • Retrained detectors on held-out store footage: cup precision 0.875 → 0.992 (recall 0.66 → 0.845), milk precision 12% → 83%, hairnet false flags 3.7% → 0.2%.
  • SheGrows, Arabic RAG (FAO, Egypt): built the Egyptian-dialect query rewriting, Arabic hybrid search, confidence routing and expert feedback loop of a WhatsApp assistant that has answered 15K+ farmer questions in production.
  • LegalTech (a ministry): built an air-gapped pipeline for scanned Arabic legal files with citation-gated retrieval. Cut OCR time per line 45× (86 s → 1.9 s, identical output) on a 4 GB GPU; caught and filtered OCR that invented text at higher confidence than real Arabic.
AI Evaluation Engineer · Freelance
Originator · United States · Remote
Jun 2026 – Jul 2026
  • Built ~40 agentic evaluation tasks testing Anthropic's Claude Sonnet and Opus against real client software, so clients can ship agents built on them. ~80% acceptance rate. Also validated tasks for Qwen3 for a second client.
  • Calibrated computer-use RL (RLVR) tasks to a 20–80% pass band across three difficulty tiers: compound-step difficulty design, small-sample confidence bounds before any verdict, and run budgets that absorb ~30% infra failures. Deterministic Python graders on the app's final state.
Data Science Instructor
RCH International · TeachCode · GOMYCODE · Tunisia
Mar 2025 – Present
  • Teach the AI Full Pack, Python through Transformers: 3 cohorts, ~50 students, mostly working professionals. In English, French and Tunisian.
Data Scientist
Qualipro® by Imagine Human · Tunisia · Hybrid
Oct 2025 – Feb 2026
  • Built a QHSE translation service in Spring AI + Groq covering 10 languages, with per-tenant caching and parallel batch translation; exposed it as an A2A agent and built a Spring AI MCP server on top.
  • Also shipped the in-app assistant and audit-automation agents. 301 automated tests, 9-stage GitLab CI with OWASP and SonarQube.
Data Scientist
D2D Analytics · Canada · Remote
Apr 2025 – Sep 2025
  • Classified 100K+ parliamentary speeches into topics; pipeline hit 95%.
  • Built a RAG system so citizens could query what was said in parliament.
AI Engineer · Internship
Wevioo (Tunis) · NTT DATA München (direct) · Hybrid
Feb 2024 – Dec 2024
  • Embedded into the NTT DATA team in Munich. Led RAG research and LLM selection on AWS SageMaker.
  • Built the cloud pipelines: S3, EC2, EKS, Docker, GitLab CI.
Data Scientist · Internship
Silver Brain AI AG · Zurich, Switzerland · Remote
Jun 2023 – Oct 2023
  • Trained a RoBERTa pipeline on 1M+ German legal records: +15% downstream accuracy.
  • Automated processing: 40% faster end to end.

Skills

LLM systems RAG (hybrid search, pgvector, re-ranking), LLM agents, agent evaluation, RLVR task and grader design, computer-use agents, Spring AI, MCP, A2A, LangChain, LlamaIndex, CrewAI, OpenAI / Azure OpenAI, Groq, Claude Computer vision NVIDIA Jetson (Orin NX / Nano), DeepStream, TensorRT, detector and classifier training, dataset curation, held-out evaluation NLP / ML Arabic and dialect NLP, Arabic OCR, Transformers (HuggingFace), RoBERTa fine-tuning, classical ML and forecasting Cloud / MLOps AWS (SageMaker, S3, EC2, EKS), Azure, Docker, Kubernetes, GitLab CI/CD, Vercel Languages & APIs Python, Java (Spring Boot), TypeScript, FastAPI, PostgreSQL, MongoDB, Redis

Side projects

Qirat (formerly Signal Desk) · solo build, live
Rule engine + LLM briefing and streaming chat · FastAPI · React · backtested on 35K+ coin-days
RAG and MCP with Spring AI · open source
Spring AI · Model Context Protocol · Java

Education & certifications

National Engineer Degree · Data Science
ESPRIT School of Engineering · 2025
Maths & Physics · preparatory cycle
Tunis Preparatory Engineering Institute · 2021
NVIDIA DLI · 4 certifications
Deep Learning · Transformer-Based NLP · Anomaly Detection · CV for Industrial Inspection