Moaripanoraune - How we built AI for market narrative detection

Inside our work

A glimpse into how we combine text, price, and flow data to build explainable AI tools for financial market research.

Research team reviewing AI outputs linking market narratives to price and flow data on multiple screens

Annotated timeline showing narrative shifts aligned with market price movements and flow changes

Close up of analyst examining text based narrative clusters on a laptop in a modern office

Whiteboard with hand drawn diagrams of AI pipeline connecting text, price, and flow data sources

Team workshop discussing governance and model explainability for financial market AI tools

Night view of Dublin city office buildings representing Irish base for AI financial research

We treat AI for financial market research as infrastructure, not magic. The models are only useful if a sceptical analyst can trace a signal back to specific inputs, stress it against history, and still decide it is worth using in a live process.

We built Moaripanoraune after years of watching narrative shifts get described only in hindsight. A central bank comment would be called a turning point, but nobody could show, with data, how that language diffused through coverage, how positioning adjusted, and how price and flow responded across assets. Our response was to assemble text, price, and flow data into one pipeline, then force our models to output intermediate labels that a human can inspect. That way, when a narrative label flips, we can show the underlying sentences, the affected instruments, and the time window, not just a headline score.

We doubt any model that claims to see the entire market in one view. So we design for modularity. One set of models focuses on classification of narrative themes in curated text sources. Another set looks at regime changes in price and flow. A third layer tests relationships between them under different historical windows. When those relationships break, we flag that explicitly, instead of smoothing it away. The result is a system that sometimes says we do not know, which is often the most honest output available in stressed markets.

We resolve the tension between automation and control by keeping humans in the loop at every critical decision. Analysts can override labels, adjust thresholds, and exclude sources that do not meet internal standards. Every change is logged. Over time this produces an internal history of how your team’s narrative framework evolved. That history becomes a resource in itself, letting new joiners see past debates and helping governance teams review why certain calls were made when they were.

We challenge the idea that market narratives are vague or purely subjective by forcing them into explicit labels, testable relationships, and written reasoning that can be revisited when conditions change.

What Moaripanoraune is and is not

We are transparent about what Moaripanoraune does not do. That is as important as what it does.

Moaripanoraune does not promise outcomes. It does not predict price paths with certainty. It does not remove the need for judgement. What it does is classify and track market narratives using text, price, and flow data, then present those classifications in a form your team can interrogate. You still carry responsibility for how you interpret and act on that information, and for how it fits into your broader risk framework.

We avoid language that suggests any tool can remove uncertainty. Financial markets remain subject to shocks, structural changes, and behavioural dynamics that no model can fully anticipate. Our stance is simple. Use AI to widen the set of patterns you can see. Use history to stress those patterns. Use governance to decide when to trust them. And always remember that past performance does not guarantee future results, even when a relationship has held for a long time.

We maintain our infrastructure and models from Ireland, with a focus on data protection and auditability. That means clear logging, controlled access, and regular reviews of data sources and model behaviour. Where required, we work with clients to align our processes with their internal standards and local regulation. We do not handle client funds, we do not execute trades, and we do not position ourselves as a substitute for independent professional advice.

Who we are

We speak as users first and builders second. We were running discretionary and systematic processes and still could not explain many price moves in a way that satisfied internal challenge. So we built AI tools that force us to connect narrative labels in text with observable patterns in price and flow, instead of relying on loose stories after the fact.

We sit at the intersection of quantitative research, natural language processing, and market microstructure. Our team includes data scientists, former sell side analysts, and a lead engineer focused on model reliability. Together we design workflows that make narrative detection auditable, so risk teams, compliance, and front office users can review the same evidence set and argue on substance.

AI research team in Ireland reviewing market narrative models

Our philosophy on AI and market narratives

We question simple stories about markets and about AI, and we design Moaripanoraune so that those questions become part of your regular research process, not an afterthought.

Assume incomplete stories, document the gaps

We start from the assumption that our first interpretation of any narrative shift is probably incomplete. So we insist on explicit labels, clear hypotheses, and repeatable tests. We compare narrative clusters with price and flow regimes across multiple windows, document when relationships hold, and record when they break. Over time this produces a catalogue of patterns and exceptions that helps teams avoid overfitting their stories to recent moves.

Treat models as living, not final

We view every model as a draft. Market structure evolves, language changes, and data sources drift. Instead of chasing a single perfect specification, we run competing models, monitor their behaviour, and retire or revise them when performance degrades. This rolling review process is written down, with reasons for each change. That way, when someone asks why a signal disappeared or shifted, there is a concrete answer, not a vague reference to market conditions.

Make explainability a shared requirement

We design Moaripanoraune so that analysts, risk managers, and compliance can all see the same evidence. That means no hidden rules, no opaque scores without supporting context, and no outputs that cannot be traced back to source material. When a narrative label moves, you can inspect the underlying text snippets, the affected instruments, and the relevant time windows. This shared view reduces friction between teams and keeps debates focused on substance, not missing data.

Separate pattern detection from decision making

We keep a strict boundary between analytical outputs and decisions. Moaripanoraune surfaces patterns in text, price, and flow data. Humans decide what, if anything, to do with that information. We remind clients that results may vary and that past performance does not guarantee future results. We do not provide personal recommendations or execute transactions. This separation protects both sides and keeps accountability where it belongs.

Plan for breakdowns, not just best cases

We build with failure modes in mind. We ask how models might misclassify narratives, how data outages could distort signals, and how regime shifts could make past relationships irrelevant. We then design alerts, fallbacks, and review processes to handle those cases. By planning for breakdowns, we reduce the temptation to overtrust clean backtests or neat visualisations, especially in stressed markets when pressure to act is highest.

We assume you already have a process for understanding market dynamics and resource allocation; our role is to add structured narrative data to that process, not to define it for you.

How teams work with our AI models

We position Moaripanoraune as a tool for teams that already question their own views. If you want confirmation, our models will frustrate you. If you want structured disagreement, they will fit.

In practice, using Moaripanoraune means feeding in your preferred text sources, mapping them into narrative clusters, and then comparing those clusters with price and flow behaviour. For example, you might look at how often a specific policy theme appears in central bank speeches, how that frequency lines up with changes in yield curves, and whether flow data shows consistent rebalancing patterns around those inflection points. Our system automates the mapping and measurement, but you still decide which sources and instruments matter for your mandate.

We also recognise that internal stakeholders care about different horizons. Short term desks may focus on intraday reactions to narrative shocks. Longer horizon teams may care about slow shifts in framing around inflation, growth, or policy credibility. Our models handle both by letting you adjust aggregation windows and thresholds. You can ask whether a narrative shift is a brief spike in language or a persistent change in how the topic is framed, and see how that aligns with observed behaviour in your chosen markets.

We work with compliance and risk functions to ensure that narrative detection outputs are documented and appropriately caveated. We make clear that results may vary and that past performance does not guarantee future results. We do not provide personal recommendations, and nothing in our outputs should be treated as tailored advice. Instead, think of Moaripanoraune as a structured lens on publicly available information and market behaviour, designed to support, not replace, your existing research and decision processes.

Why this AI market narrative detection tool exists

We built Moaripanoraune because our existing tools could not explain why the market moved when it did. Price series told us what happened. Order flow showed who reacted. News feeds and transcripts hinted at the story, but nothing tied them together in a way we could challenge. So we designed AI models that classify and predict market narrative shifts by combining text, price, and flow data into one research workflow we can actually audit.

We started as market practitioners who were tired of black box dashboards. We needed to see how a specific speech, data print, or order imbalance linked to a change in the dominant narrative. That meant building models that expose intermediate signals, versioning every experiment, and keeping a written trail of why we trusted or rejected a given feature set in live use.
How teams use us
Governance and limits: We operate from Ireland, work with compliance from the start of each engagement, and treat explainability as a non negotiable. That means clear documentation, human review before deployment, and regular recalibration as data sources or regimes change. Results may vary. Past performance does not guarantee future results. We do not provide personalised advice.

What we stand for

These values shape how we design, deploy, and maintain Moaripanoraune as an AI tool for financial market research teams.

Clarity first

We value precision in how we describe market narratives. That means defining terms, tagging sources, and resisting vague language. When we say a narrative shifted, we can point to the sentences, timestamps, and instruments involved. This discipline helps teams avoid retrofitting explanations after the fact and supports clearer internal debate.

Human control

We balance automation with human oversight. Our models scan large volumes of text, price, and flow data, but analysts can review, override, and comment on outputs. Every change is logged. Over time this creates a documented trail of how your internal views evolved, which helps governance and onboarding more than any static manual could.

Documented practice

We write things down. Model choices, data source changes, and parameter tweaks are all recorded. When a signal behaves differently, you can trace back through that record instead of guessing. This habit turns Moaripanoraune from a single tool into part of a wider research memory, shared across teams and across time.

Responsible use

We take data protection and responsible use seriously. Operating from Ireland, we align our processes with relevant regulations and client standards. We do not handle client funds or execute trades. We focus on analytical outputs only, and we encourage clients to treat those outputs as one input among many in their own frameworks.

Honest limits

We are explicit about limitations. Results may vary. Past performance does not guarantee future results. Models can miss regime shifts or misclassify narratives, especially around rare events. By stating these limits clearly, we aim to reduce overconfidence and keep attention on continuous review rather than one off deployments.

Collaborative work

We design for collaboration across research, risk, and compliance teams. Shared dashboards, common labels, and consistent documentation help different functions discuss the same narrative signals without translation overhead. This reduces friction and makes it easier to challenge assumptions before they turn into large positions or entrenched views.