Over Ikrame
Je transforme vos besoins métier en solutions IA concrètes, de A à Z.
Je m'appelle Ikrame, ingénieure IA spécialisée dans les LLMs et la donnée.
J'accompagne les entreprises de A à Z : du diagnostic de votre besoin jusqu'à la mise en production d'une solution qui tourne vraiment.
🎯 Ce que je fais concrètement :
Chatbots RAG : votre assistant IA qui répond à partir de vos propres documents (PDF, Word, Excel, emails…)
Fine-tuning LLM : un modèle entraîné sur vos données métier, qui parle votre langage
Extraction & normalisation de données : transformez vos documents bruts en données structurées et exploitables
Évaluation de modèles : je benchmarke vos solutions IA existantes et vous dis lesquelles valent vraiment le coup
💡 Pourquoi travailler avec moi ?
J'analyse d'abord votre situation avant de proposer quoi que ce soit , pas de solution cherchant un problème
J'ai de l'expérience terrain sur des projets industriels réels, pas que de la théorie
Je maîtrise toute la stack : du parsing de PDF jusqu'au déploiement en production
Je livre des solutions qui tournent
Stack : Python · LangChain · Azure · Databricks · HuggingFace · FAISS · MLflow · Streamlit · vLLM
📩 Un projet en tête ? Décrivez-moi votre problème, on trouve ensemble si l'IA peut vraiment vous aider.
Frans
Tweetalig / moedertaal
Engels
Vloeiend
Uitsluitend remote
Werkt voornamelijk remote
Werkervaring
- De Particulier à ParticulierAutomated AI-Powered Document Processing PipelineHR / PERSONEELSBELEIDjanuari 2026 - januari 2026Mulhouse, FrankrijkBuilt a fully automated end-to-end document processing pipeline using n8n as the orchestration layer, designed to eliminate manual data entry and accelerate document handling for a small business.How it works :A webhook trigger listens for new PDF documents uploaded to a Google Drive foldern8n automatically extracts the raw text from each document using a Python script executed via HTTP requestThe extracted text is sent to an LLM (OpenAI GPT-4) through a custom prompt engineered to identify and extract structured fields : client name, invoice number, date, total amount, line itemsThe structured JSON response is validated and cleaned using a JavaScript function node inside n8nValid records are automatically inserted into an Airtable database for tracking and reportingIf extraction confidence is below a defined threshold, the document is flagged and a Slack notification is sent to a human reviewer with the document preview attachedA final HTTP request triggers a confirmation email to the client via SendGrid, with a summary of the processed documentKey technical decisions :Used a multi-step prompt engineering approach with chain-of-thought reasoning to improve LLM extraction accuracy on heterogeneous document layoutsImplemented error handling and retry logic directly in n8n to handle API timeouts and malformed responsesBuilt a fallback mechanism routing low-confidence extractions to a manual review queue instead of failing silentlyResults :Reduced manual document processing time by ~80%Processed 200+ documents per week fully automaticallyHuman review required for less than 5% of documents
- LiebherrAI & Data EngineerMACHINEBOUWseptember 2025 - Vandaag (9 maanden)Colmar, FrankrijkDesigned and developed an end-to-end AI application for automatic extraction and normalization of equipment data from heavy machinery brochures across multiple industrial brands, replacing a fully manual process.
- Built and deployed a multi-tab web application serving as the main interface for the full pipeline.
- Processed complex document layouts including tables, technical specifications, and mixed content types using document intelligence techniques.
- Implemented an AI-based classification and semantic matching system to normalize equipment data consistently across heterogeneous sources.
- Designed a structured equipment database with automatic duplicate detection and data consistency mechanisms.
- Set up experiment tracking to monitor pipeline runs, log metrics, and manage model artifacts.
- Delivered a complete pipeline from raw document input to structured output with categorized equipment statuses.
- K-LINE Groupe LIEBOTAI Engineer — K-LINE Groupe LIEBOTARCHITECTUUR & STEDENBOUWKUNDEoktober 2024 - augustus 2025 (10 maanden)Les Herbiers, FrankrijkBuilt a production-ready internal AI assistant using a RAG architecture, enabling employees to query the company's entire internal knowledge base in natural language across heterogeneous document formats.
- Ingested and processed a wide variety of internal documents: PDFs, multi-sheet Excel files with scientific values, architectural window plans, Word files, and internal documentation, handling complex layouts and mixed content types.
- Built the retrieval pipeline using FAISS as the vector store for fast and scalable semantic search over embedded document chunks.
- Deployed NVIDIA Nemotron as the core LLM and integrated vLLM to enable real-time token streaming, significantly reducing perceived response latency by displaying generated tokens directly in the UI as they are produced, instead of waiting for the full response.
- Developed a custom HTML/CSS/JavaScript chat interface connected to the backend, with live streaming rendering to deliver a smooth, ChatGPT-like user experience.
- Managed the full pipeline from document ingestion and chunking to embedding, retrieval, and response generation.
Aanbevelingen
Wees de eerste die Ikrame aanbeveelt
Help deze freelancer om te schitteren door te vertellen hoe het is om met hem of haar te werken.
Deze freelancerprofielen matchen ook met zoekopdracht.
Agatha Frydrych
Backend Java Software Engineer
4.7
(3)
2
Baptiste Duhen
Fullstack developer
4.6
(4)
5
Amed Hamou
Senior Lead Developer
4
(2)
7
Audrey Champion
Web developer
4.3
(3)
4
Opleidingen
- Ingénieur Informatique & RéseauxENSISA2025
- Classes préparatoires aux grandes écolesLycée Raoul Follereau2022Physiques Technologies Sciences de l'ingénieur
Diploma's
- Préparer votre carrière dans l'IA générative par Microsoft et LinkedInLinkedIn & Microsoft2024