The transformation of professional coaching experience through AI demonstrates how advanced technology can revolutionize career coaching.
KEY FIGURES
15 min → 0 Reformulation time reduced to zero
90% User satisfaction rate
600 Active users on the new solution
ABOUT THE CLIENT
Jobmaker, active since 2015, has established itself as an innovative digital platform in professional coaching. A trusted partner of prestigious companies like EDF, Enedis, and Orange, Jobmaker supports employees in their career development through personalized programs.
TECHNICAL CHALLENGE
The major challenge was to transform the writing portion of coaching, traditionally time-consuming, into a smooth and efficient process. Our mission: develop an AI solution capable of maintaining quality while eliminating friction in the user journey.
AI Model Limitations
Resource management for Mistral-7B
Response time optimization
Balance between performance and infrastructure cost
GDPR compliance, security and AI ACT
User data anonymization
Fine-tuning process security
Protection of sensitive information
Training Data Quality
Building a representative dataset
Data cleaning and validation
Intensive testing before deployment
SOLUTION
Our solution is built around three major technological pillars:
Sophisticated Data Preparation The quality of an AI model relies on its training data. Our team developed a rigorous methodology to build a dataset of 1200 examples. The use of embeddings helped validate data relevance by measuring similarity between inputs and user responses, thus ensuring the quality of the training corpus.
Advanced Fine-tuning The core of our solution relies on Mistral-7B, a cutting-edge language model. Our fine-tuning process via Hugging Face was optimized for Jobmaker's specifics:
Adaptation to professional writing standards
Adherence to editorial guidelines
Preservation of Jobmaker's DNA in reformulations
Secure Deployment The technical infrastructure was deployed on RunPod.io in France, ensuring:
GDPR compliance
High availability
Optimal performance
METHODOLOGY
The project ran for two months, led by an expert team:
1 AI Engineer
1 Product Manager
1 Developer
Our development process was structured in three key phases:
Phase 1: Data Preparation In-depth analysis and anonymization of historical data, with particular attention to dataset quality and representativeness.
Phase 2: Fine-tuning and Development Iterative optimization of the Mistral-7B model, with testing cycles and continuous improvement based on user feedback.
Phase 3: Deployment and Optimization Progressive production rollout with performance monitoring and real-time adjustments.
RESULTS
The solution has radically transformed the coaching experience:
Technical Performance
Reduction in reformulation time from 15 minutes to instant response
90% user satisfaction rate
Copyright © 2025 Selego
