Fish n' bits

Architecture is the Advantage Using Manolin

Written by Manolin | Mar 25, 2026 6:33:10 PM

Manolin is an aquaculture intelligence company. Our two products (Watershed, for farm operators, and Harpoon, for suppliers) are built on a shared data infrastructure purpose-built for the biological and operational complexity of fish farming.

Manolin wins through its architecture. The data problem in aquaculture is not a shortage of sensors or software but the absence of a foundation that makes the data those systems generate trustworthy enough to model from. For years, that foundation has been missing. Every farm runs its own disconnected internal data stack with each technology vendor storing data differently. Each integration breaks when the biology does what biology does — i.e. fish move, populations split, production cycles span years across completely different environments.

Intelligence is only as good as the data underneath it. Farms have to get their data architecture right first.

Below, we detail the architectural decisions behind Manolin's data foundation: why we made them, what they make possible, and why they compound in ways that systems designed for a different purpose cannot replicate.

1. Our architectural philosophy

Aquaculture intelligence requires data infrastructure capable of representing biological reality, not just recording operational events from the farm. The foundation of this data must be continuous, not periodic. It must track populations, not just locations. And it must be clean and structured from the moment data arrives into a system, not as a downstream cleanup step that introduces delay and compounds error.

Our platform is built around three core commitments:

  • Biological fidelity: The data model reflects how fish actually live and grow — as populations with shared histories, not as cage-level records that lose continuity the moment fish are transferred, graded, or split. Every decision downstream depends on this.
  • Continuous integrity: Data integration, normalization, and cleaning happen in real time as data arrives. Intelligence built on stale or batch-processed data arrives too late and inherits the errors of every manual step in the chain.
  • Modeling readiness: The output of our architecture is a structured, normalized, population-traced dataset ready to be modeled. The precondition for every intelligence capability we build on top of it.

 

2. Architecture overview

The architecture addresses a problem that is both a data engineering challenge and a biological one. Fish farms generate data from feeding systems, environmental sensors, biomass estimation tools, lice counting cameras, health records, and production management systems. Each unique farm is running on different technology stacks, recording data in different formats, at different temporal resolutions, and using different identifiers for fish.

No single system was designed to integrate with the others. No common standard exists for how operational data should move between them. The result is that data which is individually accurate becomes collectively meaningless because there is no shared structure that allows it to be connected and reasoned about together.

Our architecture resolves this at each layer: ingestion, normalization, biological modeling, and intelligence delivery. Each layer operates continuously and independently, so that the foundation strengthens in real time rather than degrading between manual updates.

 

 

3. Continuous data integration across fragmented farm systems

Aquaculture data does not arrive in a clean stream. It arrives from hardware APIs, manual entry systems, vendor exports, and proprietary protocols. These are often inconsistent, with gaps, and with no guarantee that the same data event is recorded the same way twice. Our integration layer is built to handle this reality without requiring farms to change the systems they already run.

  • Hardware-agnostic connectivity: Manolin connects to the systems a farm already uses (feeding controllers, environmental sensors, lice cameras, ERP platforms, and production management software) regardless of the systems used. We do not require standardization upstream. We provide it downstream.
  • Resilient ingestion pipelines: Incoming data is written to a durable integration layer before processing begins. Connectivity gaps, system outages, and delayed uploads do not create permanent data loss, they are absorbed and reconciled when the connection is restored.
  • Temporal alignment: Farm data operates across five orders of magnitude simultaneously. Feeding events are recorded in seconds. Mortality is logged daily. Biomass is estimated weekly. Regulatory data is reported monthly. Production cycles span years. Our integration layer aligns these streams into a coherent timeline without flattening the resolution of any individual source.

This is not a one-time ETL process. Integration runs continuously, so that every new data point is incorporated into the unified structure as it arrives, not at the next export cycle.

 

4. Normalization at the scale of biological complexity

Raw data from aquaculture systems is inconsistent in ways that go beyond simple formatting differences. Vendors use different terminology for the same concepts. Units fluctuate. Recording conventions change when software is updated or staff turns over. A field labeled "mortality" in one system may represent daily count, cumulative count, or percentage and the difference matters enormously for any downstream calculation.

Our normalization layer resolves this systematically:

  • Semantic standardization: Incoming data fields are mapped to a consistent schema regardless of how the originating system labels them. Terminology differences between vendors are reconciled at ingestion, not patched downstream.
  • Unit and convention reconciliation: Data recorded in different units, at different reporting frequencies, or against different baseline conventions is normalized to a consistent representation before it enters the modeling layer.
  • Artifact detection and correction: Systematic recording errors (common in every farm system over time) are identified and flagged as data arrives. The models Manolin builds do not train on data that still contains the artifacts of inconsistent recording practices.

The result is a dataset that can be compared across farms, across vendors, and across time — without the silent errors that accumulate when normalization is treated as an afterthought.

 

5. Population tracing: the biological data model

This is the core architectural decision that separates Manolin's foundation from every farm management and analytics system built before it.

Existing aquaculture systems track data at the cage level. When fish are transferred between pens (for grading, treatment, or management) the data record resets. When populations from multiple pens are combined, their histories merge without any mechanism for maintaining continuity. The cage-level model is operationally convenient and biologically meaningless: it ties outcomes to locations rather than to the populations that produced them.

Manolin tracks at the population level: the biological unit that actually carries the fish health history.

  • Group identity through transfers: When a population moves between pens, the fish populations identity moves with it. Biological continuity is maintained across the full production cycle, from smolt intake through harvest.
  • Split and merge accounting: When populations are divided during grading or combined from multiple pens, the architecture tracks proportional composition. Downstream analysis knows which cohorts of fish populations contributed to a given outcome, and in what proportion.
  • Cross-lifecycle traceability: Performance data, treatment history, feed records, and environmental exposure are all attributed to the population that experienced them — not to the cage location where they happened to be recorded.

Population tracing cannot be bolted onto a cage-level system through a software update. The biological continuity of fish groups has to be the organizing principle of the data model from day one, not a feature layered on top of records that were never designed to carry it. This is the structural change that makes FCR calculations reliable, that connects treatment events to health outcomes with enough causal resolution to learn from, and that allows performance comparisons across cohorts, farms, and production generations to mean something.

 

 

6. High-integrity analytics on structured biological data

The analytics Manolin delivers are built directly on the population-traced, normalized, continuously-maintained dataset described above. It is intelligence that is only possible because the data underneath it is structured correctly.

  • Cohort performance analysis: Because fish group histories persist through the production cycle, performance metrics (growth rates, FCR, mortality patterns, treatment responses) can be compared across cohorts with confidence that the underlying data represents the same biological unit throughout.
  • Cross-farm benchmarking: Manolin's network of connected farms means that normalized, population-traced data from across the industry can be used to contextualize performance at any individual site. A farm's sea lice treatment in a given production cycle is not evaluated in isolation but is benchmarked against comparable cohorts under comparable conditions.
  • Causal analysis across data streams: Environmental conditions, feeding decisions, health events, and biological outcomes are connected within a unified data model. This makes it possible to ask (and answer) questions that cannot be addressed when the underlying data lives in separate systems with no shared structure.

Every analytical capability compounds with every new data connection. Each farm that joins our network improves the benchmarks. Each production cycle adds depth to the historical record. The architecture improves continuously, not by releasing new features, but because the data that powers it keeps getting richer.

 

7. Flexible access: farm platform and enterprise API

The same architectural foundation powers two different access modes, depending on how a customer wants to engage with the data.

  • Watershed (farm operators): Manolin's intelligence platform for farm teams. Built directly on the population-traced, continuously-maintained dataset, Watershed delivers the analysis that farm managers and biologists need, without requiring them to manage the data infrastructure that makes it possible.
  • Harpoon (suppliers and researchers): Manolin's intelligence layer for product suppliers. Feed companies, health product suppliers, and research organizations connect to the same structured dataset to evaluate how their products perform across real-world biological populations, at a scale and resolution that farm-by-farm reporting cannot provide.
  • Structured data API (enterprise): Larger organizations with their own internal analytics infrastructure can access Manolin's cleaned, normalized, population-traced dataset directly through a structured API. The architecture functions as a data layer that feeds existing internal tools — not as a replacement for them.

 

8. Built for the intelligence aquaculture needs next

The architecture is designed with the assumption that the intelligence requirements of the industry will continue to advance, and that advancing them requires a data foundation capable of supporting methods that do not yet exist at commercial scale in aquaculture.

  • Cross-generational learning: Because population histories are maintained across production cycles, the dataset deepens with time. Models built on three years of structured, population-traced data are meaningfully more accurate than models built on last month's exports. The architecture compounds in a way that cage-level systems cannot.
  • Model-ready data structure: The normalized, population-traced, temporally-aligned dataset Manolin maintains is structured for modeling from the ground up. As predictive and causal methods in aquaculture mature, the foundation is already in place to support them.
  • Open data access: Customers retain full access to their own data. The architecture is not designed to create dependence but is designed to make the data more useful, regardless of what tools are applied to it.

 

 

The foundation comes first

The internal data architecture most farms built were developed over time without a clear plan. A feeding controller added here, a lice camera there, a spreadsheet stitching the gaps between them. Each solving an immediate problem, but none designed to connect with the others. Over time for every farm, the manual steps have accumulated. The questions that matter most to farm managers (however) remain unanswerable.

Every layer of Manolin's architecture reflects a decision made specifically for the way aquaculture data behaves — biologically complex, operationally fragmented, and too important to leave unstructured. We have built the foundation the industry's intelligence has always required. The result compounds with every farm connected, every production cycle completed, and every data source added to the network.

Ready to connect wherever you farm.