Mustafa Khalil serves as a machine learning engineer at Swissquote Bank, where he operates within one of Switzerland's most heavily regulated financial environments. Beyond his technical role, he leads a community of practice for software developers interested in AI, focusing on evaluating tools and frameworks that ensure compliance with FINMA regulations and banking industry standards.

Expertise & Perspective: Khalil represents the intersection of AI innovation and regulatory compliance, bringing both technical machine learning expertise and practical experience in implementing AI solutions within strict financial sector constraints. His work involves balancing the productivity benefits of AI tools with the rigorous security and privacy requirements demanded by Swiss banking regulations.

Key Insights at Panoramai:

On Regulatory Framework Approach: « Hello, my name is Mustafa, I'm a machine learning engineer in Swisscot, also a software developer and I'm the only one without slides and also I'm the only one with a blazer. We don't wear it usually at work. It happened today with the blazer. » His opening humor belied the serious regulatory environment he navigates daily.

On Risk-Benefit Balance: « For me like as a software developer working in a bank with like highly regulated environment that finma, you know, like requires us to follow certain rules. I would like to pick up from the last slide that you brought Edgar. Productive productivity could like increase a lot but mistakes or errors or breach of data have a high cost on us. » Khalil emphasized how regulatory requirements shape AI adoption strategies in financial services.

On Data Privacy Imperatives: « This of course means like the correctness of what we develop, what we code or what AI suggests, how it's rolled out to production and whether it's like breaches the policies of like data privacy, you know, our client like CIDs or even our like code repositories because they are considered like part of our like strategic plannings or like confidential documents. » He detailed the comprehensive privacy considerations that govern AI tool selection.

On Deployment Strategy: « In terms of LLMs. We are also defining kind of boundaries between the third party providing LLMs like OpenAI or GPTs or CLAUDE or the others and evaluate also the open source LLMs that we can use. » Khalil described Swissquote's sophisticated approach to LLM evaluation and deployment.

On Local Infrastructure: « We kind of developed boundaries between certain areas like certain code bases or pages or documents where maybe we could use the provide like third party provider LLMs. But in general we kind of are sticking to in premise deployed LLMs as I said, for the security reasons. » He outlined the bank's hybrid approach, using local deployment as the default with carefully controlled exceptions.

On Developer Experience Enhancement: « And also to enhance the experience of developers to use them locally or use them in premise, try to try to deploy for example their instances of LLAMA CPP for example, in their machines or in some GPU shared machines somewhere. » Khalil demonstrated how regulatory constraints drive innovative technical solutions for developer productivity.

On Human-Centric Principles: « Our objective is for AI to be augmenting people's productivity, not replacing people. That's like as a principle. » He articulated Swissquote's philosophical approach to AI integration, emphasizing human augmentation over replacement.

On LLM Limitations: « I use LLMs on daily basis for my coding tasks. Like all of you here, they are still not to the level that I would assume, like a software developer that would be able to do the tasks autonomously without, like, just let them do it and go. » His honest assessment provided realistic expectations for AI capabilities in production environments.

On Context Optimization: « I had a case, I don't know if I have time I can talk about it, but I had a case where 7B model managed to solve like a complicated thing when I gave it a good context, like I provided the JIRA ticket that is related to it, the Confluence page, the page from Stack Overflow and some code and I think in 4K context, 4K tokens context, it was impressive, it found the issue and this thing failed with bigger models when I didn't provide the correct context. » Khalil's technical insight demonstrated how proper context curation can make smaller models outperform larger ones.

On Quality Assurance: « When it comes to the AI code, we'll apply the same rules that we apply to software developers, that we should have a very rigid pipeline of deployment where the code like goes from development into production... Human in the loop is very important for that to give the. » He emphasized that AI-generated code must pass the same quality gates as human-written code.

On Professional Development Cycles: « I was in Berlin in 2022, in June in we are developers conference and back then the founder of GitHub he kind of announced the GitHub copilot the first versions... And I have the same thing. I'm impressed, then I'm scared, then I'm disappointed. Something like this happens. » Khalil provided candid insight into the emotional journey of AI adoption in professional development.

On LLM vs. Human Developers: « Actually this goes to the definition of a junior employee, you know, like a junior employee is you don't expect them to do only Junior tasks, but also to develop themselves to become more than juniors. And that doesn't apply to LLM. You know, LLM. I expect the same thing from it every time. » His analysis highlighted fundamental differences between AI tools and human team members.

Strategic Positioning: Khalil represents the cautious but progressive approach to AI adoption in heavily regulated industries. His methodology involves systematic evaluation, controlled deployment, and continuous monitoring to balance innovation with compliance requirements.

Technical Philosophy: He advocates for treating AI tools as sophisticated but consistent assistants that require careful governance and context management to deliver value within regulatory constraints. His approach emphasizes augmentation over automation and human oversight over autonomous operation.

Industry Impact: Through his leadership of Swissquote's AI community of practice, Khalil influences how financial institutions across Switzerland approach AI tool evaluation and implementation. His frameworks for balancing productivity gains with regulatory compliance provide templates for other regulated industries navigating similar challenges.