Titulo Estágio
Responsible AI and Ethics in LLMs
Áreas de especialidade
Sistemas Inteligentes
Local do Estágio
Coimbra
Enquadramento
As LLMs scale into customer-facing workflows, ensuring they behave safely and reliably is not optional — it’s a requirement. However, applying guardrails introduces real tradeoffs, particularly in latency-sensitive systems. This internship will focus on designing and prototyping a guardrails framework that monitors, evaluates and enforces responsible AI behavior. The outcome will help Intermedia build internal tools for detecting and addressing unsafe or off-brand outputs, reduce risk exposure and inform governance decisions.
Objetivo
- Design and prototype advanced guardrails for input/output validation in LLM-powered features
- Benchmark tools such as NeMo Guardrails, TruLens, Llama Guard, and others
- Develop policy enforcement logic (baseline, account-level, feature-level)
- Implement observability dashboards and risk scoring systems
- Evaluate latency, cost, and user impact tradeoffs of guardrail usage
- Provide practical guidance for risk-informed AI development and deployment
Plano de Trabalhos - Semestre 1
- Onboard to the Intermedia AI Platform and familiarize with current LLM usage
- Research and benchmark guardrail tools (NeMo, TruLens, Llama Guard, etc.)
- Prototype input and output validation pipelines using rule-based and LLM-based methods
- Build initial observability framework to track flagged interactions and risk levels
- Analyze tradeoffs in latency, accuracy, and UX caused by applying guardrails
Plano de Trabalhos - Semestre 2
- Expand the policy enforcement system to support different levels (e.g., per-account, per-feature)
- Implement evaluation frameworks (Phoenix, DeepEval, RAGAS) to assess guardrail effectiveness
- Prototype human-in-the-loop workflows for escalation and feedback
- Run experiments comparing filtered vs. unfiltered output performance and risks
- Deliver final presentation and documentation with practical deployment guidelines
Condições
- Monthly stipend of €900, based on a full-time commitment (40 hours per week)
- Hybrid work model: interns are expected to work from the company office 2–3 days per week
- PC or Mac for the duration of the internship
- Integration into a Product Area team, working closely with engineers and stakeholders
- A mentor will be assigned to support the intern’s development and success
Observações
What We Are Looking For:
- Self-motivated individuals with a strong desire to learn and a collaborative mindset
- Final-year Master’s students in Computer Science, Artificial Intelligence, Engineering or related fields
- A solid understanding of software architecture fundamentals
- Demonstrated experience with at least one relevant programming language or technology (e.g., Python, Java, .NET or C/C++)
- A genuine interest in software design and development
- Enthusiasm for CCaaS/UCaaS, AI/ML or LLM safety frameworks
Orientador
Crispim Tribuna
ctribuna@intermedia.com 📩