PhD position (start: early 2026): Tool-Augmented LLMs for Enterprise Data AI
Role details
Job location
Tech stack
Job description
Enterprises are turning to LLMs for high-stakes, data-driven decisions over structured data, such as databases and knowledge graphs. This PhD sets a three-year agenda to build and validate thinking-with-tools, an approach where the model embeds live tool use (e.g. SQL/SPARQL, Python) inside its reasoning. Beyond accuracy gains, the goal is an efficient decision system that knows when it's unsure, asks for clarification, and escalates when judgment matters. Consider the question: "If we cut Product X's price by 5% next quarter, what revenue should we expect?", the system reasons, pulls the right numbers, exposes its assumptions, and invites a human to confirm or adjust when needed. This aligns with broader research priorities on human-AI decision support and governance in complex settings.
We will keep an open, comparative stance: tool-augmented reasoning is a promising solution, not a foregone conclusion. A core part of the PhD is to demonstrate, on company data and realistic scenarios, when it surpasses agentic pipelines (LLMs orchestrating tools step-by-step) in terms of quality of the solutions, efficiency, costs and their trade-offs.
You will design the framework, fine-tune open LLMs to plan and run safe, efficient tools, and develop policies that balance accuracy, cost, and latency. This includes building a sandbox reflecting enterprise schemas; craft training traces that interleave reasoning and tool calls; and prototype realistic Aily Labs use cases (e.g., forecasting and customer insights). Examples of possible research outcomes are: a cost-aware reasoning optimizer that learns when/how to invoke tools; typed, auditable traces and human-in-the-loop mechanisms (uncertainty, clarification, escalation); and budget-aware routing that adapts across schemas and tasks. Impact will be measured with clear evaluations of accuracy/latency/cost trade-offs, robustness to schema evolution, and governance considerations.
Requirements
Do you have experience in SQL?, Do you have a Master's degree?, Environment. In this industrial PhD you'll be hired by Aily Labs and embedded with product/data teams in Barcelona/Madrid (Spain), while enrolled at EURECOM under Prof. Paolo Papotti supervision. Duration 36 months; working language English.
Profile. Strong background in ML/NLP and databases (SQL/DBMS/KGs); solid Python and experimentation. Experience with LLM fine-tuning, text-to-SQL, tool use/agents, or RL is a plus.
