TransitionCV: AI-Powered Resume and Cover Letter Assistant

The Art of Prompt Engineering

At the core of TransitionCV’s success is prompt engineering—the process of designing, testing, and refining instructions to guide the AI's behavior and responses effectively.

Prompt engineering is vital because it shapes how the AI understands and responds to user inputs. For a career-focused tool like TransitionCV, we needed the AI to deliver high-quality, contextually accurate outputs tailored to the Danish job market. This required iterative experimentation and learning to find the right balance between general applicability and localized expertise.

Here’s how we approached it:

• Iterative Testing and Learning: Crafting effective prompts required continuous testing with real-world examples. Each iteration allowed us to identify gaps, improve clarity, and refine outputs to align more closely with user needs.
• Domain Expertise Integration: We worked to embed the knowledge of Nordic Transition and industry norms into the AI’s responses. This involved feeding the AI carefully structured prompts that reflect local expectations, job market terminology, and cultural nuances.
• Balancing General and Specific Needs: While focusing on Danish users, we designed the system to remain adaptable for future expansion, like supporting other languages and markets.

Prompt engineering isn’t a one-time task; it’s a dynamic process of learning, tweaking, and optimizing. By investing in this iterative process, we ensured that TransitionCV provides users with actionable and relevant advice, making it a trusted companion in the job application process.

How Did We Build It?

TransitionCV is built with Next.js 14 and hosted on Vercel for seamless performance. API route handlers are utilized for serverless calls to the Azure OpenAI endpoint.

Why Azure OpenAI?
It provides a secure, scalable environment with strong data protection measures. Additionally, it offers flexibility, allowing us to switch between large language models (LLMs) or scale functionalities easily.

Privacy Matters

We take your privacy seriously. TransitionCV does not store your resume or any input you provide. The system only keeps track of usage statistics, such as the number of CVs generated, without using cookies or storing visitor data.

For AI processing, we rely on Azure OpenAI Service, which operates within Microsoft’s secure Azure environment. Importantly, this service does not interact with OpenAI’s public APIs, ensuring that your data remains private and secure.

What’s Next?

Through this journey, we’ve gained significant insights into prompt engineering and fine-tuning ChatGPT. However, the biggest obstacle isn’t the AI itself—it’s the accuracy of the data. Fortunately, Nordic Transition’s flagship product, Career Builder, excels in this area. Not only it has candidates’ CVs, but it also maps their ambitions and aspirations for their personal careers.

Recognizing this strength, we’ve shifted our focus to leverage this data in Career Builder by integrating Retrieval-Augmented Generation (RAG) and agent-based AI services using Microsoft Semantic Kernel. This approach enables us to train the model with highly accurate data, paving the way for a personalized AI assistant that truly understands candidates context.

Stay tuned for more updates!