AI Engineer
Pubblicato il 21-10-2026 - Nebuly in Piemonte
About Nebuly
Nebuly is a fast-growing, VC-backed startup building the data and intelligence infrastructure for enterprise AI agents.
Every enterprise function — sales, customer support, HR, finance — is deploying AI agents at scale. Every day, millions of conversations happen between employees and customers and these agents. Those conversations are the most valuable new data asset an enterprise produces — but they are unstructured, high-volume, and invisible without the right infrastructure.
Nebuly plugs into the conversations flowing between users and enterprise AI agents, builds structured knowledge graphs from them, and surfaces the patterns that help leaders make decisions on adoption, quality, and ROI.
Every new type of enterprise data has created a new category of software. CRM for customer interactions. Product analytics for user behavior. Nebuly is building that infrastructure layer for AI agent conversational data — a category that will touch every enterprise on the planet.
Our customers include Vodafone, CNH, Oura, and others. Demand is accelerating fast — and we are building the team to match.
What you’ll do
The AI team sits at the core of Nebuly’s product: every customer-facing insight — from taxonomy classification to failure detection to ROI estimation — is powered by the models and pipelines this team builds and operates. You’ll work closely with the Full Stack and Product teams to ship AI-driven features that go straight into production for enterprise customers, and you’ll be one of the people directly responsible for the accuracy, reliability, and scalability of Nebuly’s core AI systems.
This is a hands-on role. Over just the last few weeks,
the AI team has: extended the classification rules framework and taxonomy system to support multi-label classification, advanced user-group logic, and configurable categories; shipped core pieces of our new ROI product (cost-per-hour-per-class mapping and trust scoring amongst others); improved PII model robustness; built out our MCP server and its discovery capabilities; stood up a unified automated testing framework for AI models and consolidated testing; worked on LLM inference efficiency (batch routing for prefix-cache hit rate, API token usage determinism, GPT model migrations); shipped a local/browser plugin to capture AI agent behavior (e.g. Claude, Cowork); improved classification and data quality across accounts like Ferrari, Iveco, Fastweb, Oura, and Moncler.
Responsibilities
- Design, train, and iterate on classification and taxonomy models (including multi-label classification and rule-based/LLM-based hybrids) that power Nebuly’s core product
- Build and extend the ingestion and enrichment pipelines that process enterprise customer interaction data at scale, with attention to performance, cost, and reliability
- Work on our ROI product line: estimating human time saved, mapping cost per hour per class, and building the taxonomy and trust logic behind it
- Improve the robustness and performance of PII detection models, including evaluating alternative open-source LLMs and improving training data
- Contribute to and extend the MCP server and other AI-facing tooling used internally and by customers
- Build and maintain a shared automated testing framework for evaluating models, datasets, and classification pipelines
- Investigate and fix model and data issues reported directly by enterprise customers, often under production time pressure
- Work on LLM inference efficiency: batching, caching, token usage, and model migrations as providers evolve
- Run PoCs and technical discovery for new AI-driven product capabilities, from scoping to prototype
- Collaborate closely with Product and Engineering to translate ambiguous product requests into shipped AI features
- Participate in code and model reviews, and help raise the technical bar of the AI team
Requirements
- Minimum 1-3 years of hands-on experience building and shipping ML/AI systems in production
- Strong Python skills, including experience with data processing and ML/LLM pipelines
- Practical experience with classification systems (traditional ML and/or LLM-based approaches, prompt-based or fine-tuned)
- Comfort working with messy, real-world data and debugging model behaviour on live customer data
- Experience with at least one of: PostgreSQL, ClickHouse, or similar data stores
- Ability to write tests and think rigorously about evaluation and validation of ML systems
- Strong problem-solving skills and the autonomy to take a project from ambiguous discovery to shipped feature
- Comfo