Modena, 22 December 2025.
The 2024 edition of this letter sparked a remarkable amount of engagement: not only inbound questions and conversations, but also a wave of internal discussions, ideas, and proposals on how we could push ourselves further as a tech-first company. It became clear that taking the time to reflect publicly on our progress has real value, both inside and outside Axyon AI.
So here we are, one year later, looking back at what we achieved in 2025. The act of writing and distilling our work into a few key highlights is, again, an incredibly (hard and) fruitful exercise. I genuinely encourage any tech leader to do the same periodically: it forces clarity, reveals patterns, and surfaces achievements that would otherwise be lost in the noise of everyday delivery.
This year has been intense, in the best possible way. Across 2025, we completed 55 projects1, spanning 24 business-originated initiatives and 21 tech-driven ones. The selection below is a very small sample of what the team built and improved, but each of those items represents a meaningful step forward in our mission and in the maturity of our tech stack.
Here are the three2 highlights I’m most proud of.
1. Axyon Lens: Unlocking True End-to-End Explainability
- What: “The explainability of a Machine Learning (ML) model can be as crucial as its performance” has been a foundational mantra at Axyon AI for many years. In 2025, we set out to take explainability to the next level. Beyond traditional feature-based explanations, Axyon Lens provides a unified framework for end-to-end explainability, from high-level strategy performance analysis and attribution to a detailed examination of model predictions, feature importance, dependency and interaction analysis, all the way to individual input time series.
- Why it matters: Lens gives investors genuinely transparent clarity. Let’s consider a simple question like “What changed in Q1 2025 in this strategy, and why?”. Before Axyon Lens, we could reply: “The underlying model attributed more importance to context data”. With Axyon Lens, we can say “The model decreased its exposure on Tech & Industrials in favour of Utilities, because it started preferring low beta stocks, given the broader financial and economic context, namely higher FX volatility and slowing global trade, as indicated by dropping freight earnings3”. This level of interpretability strengthens client trust, supports due-diligence and compliance workflows, and sharply differentiates Axyon AI in a market where transparency is increasingly mandatory. Axyon Lens is becoming a cornerstone of our value proposition: institutional-grade explainability, built into every model and every strategy.
2. Pioneer: A Unified API Backbone for Axyon IRIS
- What: As Axyon IRIS grows, so does the complexity of data and signal exchanges among internal components and services. To address this, in early 2025 we designed and built Pioneer, a new Django-based REST API that serves as the authoritative data access layer for the whole Axyon IRIS ecosystem. After months of production hardening under internal load, Pioneer is now open to external consumption as well, providing a fast, standardised, scalable, and secure interface for clients and partners.
- Why it matters: Pioneer is a foundational upgrade to our architecture: across IRIS, services no longer need to depend on one another directly, only on a well-defined API contract. This reduces integration effort across the board, isolates upgrades, simplifies testing, and increases platform reliability. For clients, it means modern, stable, industry-standard programmatic access to IRIS outputs. For the engineering team, it is a major maintainability and scalability leap that will pay long-term dividends.
3. Axyon Foresight: An Agentic AI Platform for Ultra-Fast Thematic Investing
- What: Thematic investing has historically been a slow, manual, and backward-looking process: turning an investment thesis into a screened universe can take weeks or months, requiring endless debate over definitions, hand-crafted criteria, spreadsheet scoring, and subjective judgement. In 2025, we4 set out to change this entirely with the release of Axyon Foresight, our new agentic AI engine for theme discovery, research, and universe screening. Starting from a high-level theme definition, Axyon Foresight automatically generates a complete KPI framework, retrieves and interprets evidence from wide data sources, evaluates thousands of stocks, and produces a fully explainable investable basket in a matter of hours. But there’s more: Axyon Foresight also includes a “discovery” engine, which continuously scans the markets and news for potentially new themes to evaluate. At every step, humans retain oversight: candidate themes, KPIs, weights, assumptions, and interpretations can be refined iteratively before the final basket is constructed. This human-in-the-loop workflow, combining multi-agent reasoning, structured evaluation, and transparent documentation, allows a user to transform a vague thematic idea (“Energy Storage”, “Space Economy”, or “GLP-1 Agonist Drug Adoption & Second Order Effects on Consumer Behaviour”5) into a measurable, data-driven, auditable screening process with unprecedented speed and consistency.
- Why it matters: Foresight unlocks a new frontier for investors: forward-looking thematic construction at scale, built on a foundation of auditability and explainability. Traditional thematic products rely heavily on intuition, narrative bias, or static classifications; Axyon Foresight replaces this with a systematic yet flexible framework where every stock selection has a documented rationale, every KPI has a traceable methodology, and every decision is grounded in evidence. Clients can ideate and launch thematic products in days rather than months, explore emerging trends rapidly, challenge or refine the model’s reasoning, and integrate the outputs directly into their investment workflows.
So many other exciting projects didn’t make this shortlist, yet were absolutely foundational to the evolution of our tech stack - from the complete overhaul of our backtesting engine, to continuous data and modelling research, the adoption of serverless computing at scale, countless enhancements to observability, and an ongoing, tireless push toward automation and testing.
All of this is a testament to our teamwork, navigating a massive flood of Jira tickets, deciphering cryptic whiteboard sketches6, enduring endless debugging sessions, taming truly outlandish edge cases, and celebrating those late-night breakthroughs that make it all worthwhile.
Here's to another year of learning fast, building boldly, and pushing the boundaries of what’s possible at Axyon AI.
Best regards,
Jacopo Credi
Chief Technology Officer & Co-founder
1 At Axyon AI, a project (sometimes also referred to as an “Epic” by the Jira lovers) is an initiative with a responsible team, a timeline extending for more than a two-week sprint, and clearly defined deliverable.
2 It is a universally acknowledged truth that any decent list must strictly contain either 3, 5, 10, or 101 items.
3 This is not made up, but part of a real case study - available upon request.
4 A *huge* shout-out to Daniele Grassi, our CEO. He actually built the initial version of Axyon Foresight in his own time and is still the top contributor today!
5 Fun (?) fact: this is a theme I personally came up with, feeling quite proud of myself… until I realised that Axyon Foresight’s theme discovery agent had already identified it one month earlier!!
6 One’s own especially - it was so clear yesterday.