The Research Infrastructure
behind Integrative Health
Cynara scolymus
Artichoke (Cynara scolymus) contains chlorogenic acid, cynarin, and luteolin-7-O-glucoside that stimulate bile secretion and enhance gallbladder emptying. These polyphenols inhibit HMG-CoA reductase, reducing total and LDL cholesterol in hyperlipidemic patients. The sesquiterpene lactone cynaropicrin activates Nrf2 pathways and increases hepatic glutathione, providing antioxidant protection. Artichoke extracts reduce malondialdehyde while boosting superoxide dismutase, catalase, and glutathione peroxidase in damaged hepatocytes. The compounds inhibit NF-κB signaling and decrease TNF-α and IL-6 expression. Inulin-type fructans act as prebiotics, selectively promoting beneficial gut bacteria growth.
Passiflora incarnata
Passiflora incarnata contains benzoflavone derivatives including chrysin and isovitexin that bind to benzodiazepine receptors, producing anxiolytic effects without significant sedation or cognitive impairment. The harmala alkaloids harmine and harmaline inhibit monoamine oxidase A, increasing serotonin and norepinephrine availability in the central nervous system. Maltol and ethyl maltol act as GABA receptor modulators, enhancing chloride channel conductance and producing mild sedative effects
Valeriana officinalis
Valeriana contains valerenic acid and acetoxy valerenic acid that act as positive allosteric modulators of GABA-A receptors, enhancing inhibitory neurotransmission without direct benzodiazepine-site binding. The sesquiterpenes valerenal and valeranone inhibit GABA transaminase, reducing GABA catabolism
Hypericum perforatum
Hypericum perforatum contains hypericin and pseudohypericin, naphthodianthrones that inhibit dopamine β-hydroxylase and modulate multiple neurotransmitter systems including serotonin, norepinephrine, and dopamine reuptake. The phloroglucinol derivative hyperforin acts as a potent reuptake inhibitor by elevating intracellular sodium concentrations, indirectly blocking monoamine transporters SERT, NET, and DAT. Flavonoids including hyperoside, rutin, and quercetin demonstrate MAO-A and MAO-B inhibition, reducing monoamine degradation
Mackerel
Scomber scombrus provides eicosapentaenoic acid (EPA) and docosahexaenoic acid (DHA), omega-3 polyunsaturated fatty acids that modulate inflammatory responses by competing with arachidonic acid for cyclooxygenase and lipoxygenase enzymes, reducing pro-inflammatory eicosanoid synthesis
Silybum marianum
Silybum marianum contains the flavonolignan complex silymarin, composed primarily of silybin, silydianin, and silychristin, which stabilize hepatocyte membranes by inhibiting lipid peroxidation and preventing toxin penetration. Silybin competitively inhibits hepatocyte uptake of amatoxins and phalloidin by blocking organic anion transport proteins. The compound stimulates RNA polymerase I activity, enhancing ribosomal protein synthesis and accelerating hepatocyte regeneration following toxic injury. Silymarin increases hepatic glutathione concentrations by upregulating γ-glutamylcysteine synthetase expression and enhancing cysteine availability.
Ginkgo biloba
Ginkgo biloba contains flavonoid glycosides including quercetin, kaempferol, and isorhamnetin that scavenge reactive oxygen species and inhibit lipid peroxidation in neuronal membranes. The terpene lactones ginkgolide A, B, C, and bilobalide act as platelet-activating factor antagonists, reducing platelet aggregation and improving microcirculatory blood flow.
Lion's mane
Hericium erinaceus contains erinacines and hericenones, diterpenoid and aromatic compounds that stimulate nerve growth factor synthesis in astrocytes and promote neuronal differentiation and axonal outgrowth. Erinacine A crosses the blood-brain barrier and induces NGF gene expression through activation of the JNK signaling pathway. The β-glucan polysaccharides demonstrate immunomodulatory effects by activating macrophages, natural killer cells, and enhancing cytokine production including IL-1β, IL-6, and TNF-α.
Brassica
Brassica oleracea contains glucosinolates including glucoraphanin, glucobrassicin, and sinigrin that undergo myrosinase-catalyzed hydrolysis to produce bioactive isothiocyanates and indole-3-carbinol upon tissue disruption. Sulforaphane, derived from glucoraphanin, potently induces phase II detoxification enzymes by activating the Keap1-Nrf2-ARE pathway, enhancing glutathione S-transferase, NAD(P)H quinone oxidoreductase, and UDP-glucuronosyltransferase expression
Our Research Foundation Commitment
In an era of information overload and unsubstantiated wellness claims, rigorous research methodology forms the cornerstone of responsible nutrition practice. Our vision integrates a clinical research intensive approach, where botanical wisdom converges with cutting-edge analytical methods, orthomolecular protocols, and empathy.
For patients navigating chronic pain, cancer, metabolic dysfunction, digestive disorders, or seeking optimal health through evidence-based interventions, the tools now exist to move beyond symptom management toward genuine restoration of physiological function.
Our commitment extends beyond implementing protocols to actively contributing to the evolving evidence base by clinical collaboration and scientific literature review. The transformation of clinical nutrition from empirical tradition to personalised care is not tomorrow's promise, it is today's imperative.
What Type of Research We Do?
Practical, Clinically Focused, Directly
Connected to Real-World Application.
We conduct clinical and medical reviews, as well as foods and phytochemicals assessment of therapeutic effectiveness, safety, mechanisms of action, and clinical relevance. We study biochemical components of nutrients and therapeutic compounds, including how they work in the body and how they can be used in structured protocols.
We analyse nutrient–nutrient interactions and nutrient–pharmaceutical interactions to ensure safety and coherence in integrative interventions. We document the historical and geographical use of foods, plants, and nutritional tools, connecting traditional knowledge with current scientific validation.
We continuously assess current data in mycotherapy, phytotherapy, dietotherapy, nutrition therapy, clinical nutrition, and lifestyle interventions supported by behavioural therapy strategies. This includes specific research on advances in pain management and inflammatory modulation, cancer and metabolic health.
We design and refine orthomolecular protocols based on evidence, biochemical individuality, and measurable outcomes, ensuring that research translates into clear, defensible practice, as well as helping other clinicians, therapists and researchers.
Translation of Treatises
& Knowledge Transfer
Kalavik engages in the rigorous translation of scientific volumes, monographs, clinical manuals, and specialised treatises aligned with our core domains — clinical nutrition, history of drugs, phytotherapy, history and philosophy of emotions, integrative diagnostics, and complex chronic care. Our work goes beyond literal conversion of language; it ensures conceptual precision, terminological fidelity, and epistemic coherence across disciplines. Each translation preserves methodological nuance, clinical intent, and evidentiary context, allowing advanced knowledge to circulate responsibly between researchers, practitioners, and international audiences.
Quality Standards
on AI Assisted Research
Epistemic Values
We depart from the point that a research methodology is robust when it is systematic in structure, reproducible in process, auditable in execution, and explicitly aware of bias at every analytical step. It must be transparent in its reasoning and fully traceable to primary evidence, allowing every conclusion to be verified against its source.
Artificial intelligence can significantly enhance efficiency. Not only accelerating screening and finding proper literature, but also in the synthesis and data structuring, comparison and epistemic accuracy. We also defend that methodological quality cannot be delegated to automation.
Standards must remain human-defined, protocol-driven, and ethically governed. AI precision tools support the process; they do not replace scientific responsibility nor the original design of clinical strategies. A defensible research workflow should optimise for:
1. Comprehensiveness – Did we capture all or the major relevant evidence?
2. Accuracy – Are extracted data and summaries correct?
3. Reproducibility – Can another researcher lead to similar conclusions?
4. Transparency – Are decisions and processes properly documented?
5. Efficiency – Was time/resources used optimally?
6. Bias Control – Were systematic distortions minimised?
7. Interpretive Validity – Are conclusions justified by data?
On Using AI for Research,
Sourcing & Accuracy
At Kalavik, artificial intelligence is not used to replace expertise. It is used to strengthen the structure of how evidence is gathered, organised, and interpreted. Our objective is methodological clarity: wide retrieval, structured extraction, transparent synthesis, and accountable human judgment. AI is therefore integrated as infrastructure within a clearly defined research framework that we want to explain on this page.
Comprehensive Retrieval Across
Large-Scale Academic Databases
The quality of any clinical or research conclusion depends first on the breadth and representativeness of the literature base examined.
We trust Elicit because it is a research powerhouse created for researchers by researchers. It provides access to a large multidisciplinary corpus drawn from Semantic Scholar, PubMed, and OpenAlex. Elicit's ecosystem spans hundreds of millions of publications across biomedical, clinical, and multidisciplinary domains.
Elicit searches across over 138 million academic papers from Semantic Scholar (+200 million publications across all fields), PubMed (+33 million citations on biomedical literature, life sciences, and clinical medicine), and OpenAlex (243 million publications from +260,000 sources; this is about 2x broader coverage than Web of Science or Scopus). This covers all academic disciplines.
Independent evaluations indicate that multidisciplinary databases such as OpenAlex achieve broader coverage than traditional subscription databases in comparative analyses, retrieving substantially more records than Scopus and more than Web of Science in specific comparisons (Thelwall & Jiang, 2025; Fuente-Gutiérrez & Kippes, 2025; Elicit: 'Source of papers').
In biomedical contexts, OpenAlex demonstrated 98.6% coverage of guideline-cited papers in one evaluation, ahead of EMBASE (96.8%) and PubMed (93.0). Also, 90.5% of all articles were retrievable from all four databases. All articles that were not retrievable from OpenAlex and Semantic Scholar were either assessed as medium or high RoB. In contrast, both Embase and PubMed missed articles that were of high quality, ie.: low RoB. (Rajit et al., 2025). For us, this is a methodological point: broader retrieval increases the likelihood that relevant, high-quality evidence enters the analytic process.
Structured Screening & Extraction
with Defined Protocols
Elicit’s structured data extraction has been evaluated against human-verified “gold standard” datasets and demonstrated approximately 94% agreement under that validation framework. This assessment was conducted by comparing AI-extracted fields with reference datasets that had been independently verified by human reviewers, with additional LLM-based verification applied to ensure alignment. In terms of screening performance, the system reported:
• 93.6% recall (sensitivity) — correctly identifying and screening in the majority of relevant papers.
• 62.8% specificity (true negative rate) — correctly excluding a substantial proportion of irrelevant papers.
These figures indicate strong performance in identifying relevant literature while maintaining moderate discrimination in filtering out non-relevant records.
Independent evaluations emphasise the continued necessity of human revision, filtering, and methodological oversight. AI research assistants such as Elicit demonstrably accelerate structured and repetitive components of the review process and, in defined domains, can approach the extraction accuracy of individual human reviewers. However, these systems do not replace comprehensive, critically appraised research workflows that require contextual judgment, bias assessment, and interpretive reasoning.
Evidence suggests that Elicit may perform comparably to —and in certain bounded conditions potentially more consistently than— a single researcher in scenarios such as:
• Extracting clearly standardised variables or predefined metrics directly from full-text articles.
• Operating within an explicit extraction protocol supported by carefully structured prompts.
• Processing large corpora where sustained human screening and extraction would otherwise be affected by fatigue-related variability.
Such findings support the role of AI as a structured augmentation tool within defined methodological boundaries, rather than as a substitute for expert evaluation.
Onother comparative analysis reported an overall extraction accuracy for Elicit of approximately 91%, positioning it slightly above other evaluated AI tools and within the performance range of experienced individual reviewers. Under conditions where a clearly defined benchmark dataset and extraction schema are established, the findings suggest that Elicit’s performance can align with — and in certain cases approximate or exceed— that of a single human reviewer.
The same study observed particularly strong accuracy in the extraction of standardised variables, where structured fields and predefined data points reduce interpretive ambiguity. On this basis, the authors propose a hybrid workflow model: AI-assisted extraction may substitute for the second human extractor in dual-review processes, with a human reviewer retaining responsibility for reconciling discrepancies between the primary human extraction and the AI output (Andersen, 2025). This model preserves human adjudication while leveraging AI for structured consistency in large-scale review workflows.
On Using Elicit & SciSpace
The research indicates that AI-assisted tools enable researchers to access and process very large volumes of academic literature with significantly greater speed and organisational capacity than traditional manual approaches. Rather than functioning as shortcuts, these systems reshape the mechanics of review by expanding retrieval bandwidth and reducing procedural friction in synthesis workflows.
Grounded in the Technology Acceptance Model (TAM) and Cognitive Load Theory (CLT), comparative analysis suggests that such tools are perceived as particularly valuable because they streamline literature review and evidence synthesis processes while lowering cognitive burden in repetitive or structurally defined tasks. In comparative evaluation of three systems, differentiated strengths were identified:
• Elicit demonstrated particular suitability for structured evidence synthesis and extraction tasks.
• SciSpace showed strength in writing support and collaborative research workflows.
• Consensus was especially effective in literature review and summarisation functions.
These findings support a task-aligned deployment model rather than a one-tool approach. Furthermore, early evidence suggests that AI research assistants contribute to improved contextual linkage across studies, better management of contradictory findings, enhanced domain-specific extraction accuracy in structured variables, and the capacity to integrate updates dynamically as new research becomes available.
While continued methodological refinement and responsible adoption remain necessary, the trajectory of evidence indicates that these systems can meaningfully augment —though not replace— rigorous research practice when embedded within clearly defined human-governed workflows.
On Elicit's Search Methodology
Elicit’s literature searches are conducted through an LLM-powered semantic retrieval system designed to move beyond conventional keyword-based querying. The methodological process unfolds in distinct stages.
First, an initial retrieval phase uses neural language model–based semantic embeddings to identify candidate papers according to conceptual similarity rather than simple lexical overlap. This allows the system to recognise thematic alignment even when terminology differs across disciplines or publication contexts.
Second, the candidate pool is subjected to relevancy scoring and ranking using a transformer-based classification model trained to evaluate topical alignment. Rather than relying on frequency of keyword occurrence, this stage assesses contextual correspondence between the research query and the substantive content of each paper.
Third, the top 500 papers, ranked by relevance score, are selected for deeper consideration. This thresholded selection narrows the corpus to a concentrated, high-relevance subset while preserving breadth.
Fourth, an explicit screening phase is conducted, during which papers are systematically evaluated against predefined inclusion and exclusion criteria to determine final eligibility.
This multi-stage methodology leverages large language models to interpret complex research intents that extend beyond Boolean logic or simple keyword matching. By combining embedding-based retrieval with contextual re-ranking, the approach increases both recall and precision, while enriching the proportion of substantively relevant literature within the analysed corpus compared to conventional search workflows.
On Safety
As AI becomes increasingly embedded in research and decision-making, epistemic reliability becomes central. Elicit adopts a process-supervision approach: instead of relying on opaque, monolithic models, it builds compositional systems structured around human-understandable task decompositions with supervised intermediate reasoning steps. This enables large-scale AI support while preserving transparency, inspectability, and risk control—particularly in high-impact contexts. The broader aim is to promote process-based, accountable AI infrastructures over increasingly large black-box systems, strengthening reliability in research and knowledge work.
To know more about their approach, please read Andreas Stuhlmüller’s ‘Improving Epistemics’, and Stuhlmüller & Byun's ‘Supervise Process’.
What this Means
Patients reading our knowledge entries benefit from a process that is less dependent on one person’s memory or selective citation habits, while still being clinically reviewed and documented by a human professional.
Clinicians and Therapists receive a corpus where extraction is systematic and repeatable, and where uncertainty is documented rather than hidden. It informs of history, uses, clinical practicality and interactions with a level of depth and pragmaticity beyond common corpora.
Researchers can audit the path from query > corpus > extraction schema > synthesis, which is the difference between an individual on-demand AI summary per query and a proper methodological functional infrastructure.
Kalavik’s position is simple: AI increases deep and wide results with structured consistency; humans remain responsible for meaning, safety, and clinical judgment. The point is to keep the work falsifiable, reviewable, and clinically accountable.
Our Research Method
5-Phase Quality Framework
Scientific credibility does not emerge from isolated studies or selective citations—it emerges from structure, traceability, and disciplined execution. Our research method is built on a clearly defined 5-Phase Quality Framework that governs how evidence is identified, evaluated, synthesised, and translated into The Kalavik Atlas and our practice.
This framework integrates systematic methodology, documented tool usage, performance checkpoints, and auditable outputs. From protocol registration to final synthesis, each phase is designed to ensure reproducibility, bias control, and transparent reporting. The result is not only academic integrity, but clinically actionable knowledge—structured, verifiable, and ready to inform precision interventions. Our pyramid of evidence is defined as follows:
Phase 1 – Protocol Definition
• Research question formalised – The clinical or scientific question is precisely articulated to eliminate ambiguity and anchor the entire investigation.
• Inclusion/exclusion criteria defined – Clear eligibility parameters are established to ensure consistency, relevance, and methodological discipline.
• Protocol registered – The study framework is publicly documented in advance to safeguard transparency and prevent post-hoc adjustments.
Phase 2 – Search & Retrieval
• Multi-database search – Comprehensive searches across multiple databases reduce selection bias and broaden evidentiary scope.
• Documented strings – All search strategies are recorded verbatim to guarantee reproducibility and auditability.
• Sensitivity check – Search performance is tested to confirm that key benchmark studies are successfully captured.
Phase 3 – Screening & Extraction
• Standardised extraction – Structured data summarising and comparison ensure uniform capture of outcomes, methods, and effect measures.
• Random audit – Targeted verification of extracted data reinforces accuracy and procedural integrity.
Phase 4 – Critical Appraisal
• Risk-of-bias scoring – Each study is evaluated for methodological weaknesses, profits, legal bonds or interest relationships that may distort results.
• Certainty grading – The strength of the cumulative evidence is rated to distinguish robust findings from provisional signals.
Phase 5 – Synthesis & Reporting
• Writing process - We merge information into practical knowledge.
• Explicit limitations – Methodological constraints are openly acknowledged to contextualise interpretation.
• Data sharing – Extracted datasets and analytic decisions are made accessible to enable verification and reading from The Kalavik Atlas entries.
Our Output
From Evidence to Application
Our work translates research into structured, measurable, and clinically relevant practice. Every output—whether therapeutic, educational, or collaborative—is grounded in methodological rigor and evidence traceability. Across all areas, our objective is consistent: to transform evidence into precise, accountable, and useful practice.
Personalised Nutrition Practice
Our core output is precision-based, clinically grounded, backed up by research on The Kalavik Atlas that clients can check up for free any time, impacting very positively and transparently in one-to-one work, but also increasing their knowledge making the sessions a trusted space to learn. Each protocol integrates biomarkers, phytotherapy, mycotherapy, microbiota, orthomolecular strategies, and lifestyle modulation into a coherent and continuously monitored personalised intervention. Outcomes are documented, evaluated, and refined—ensuring analytical clarity rather than anecdotal care.
The Kalavik Atlas Publications
The Kalavik Atlas is our curated knowledge base: systematically structured, referenced entries on nutrients, botanicals, phytochemicals, mycotherapy, nutritional tools such as supplements and nutrient-based substances, as well as our section for orthomolecular protocols and clinical strategies, recipes and resources. It functions as both a professional resource and a decision-support framework for nutrition/clinical practitioners and clients.
Case Studies & Research Collaboration
We produce anonymised, structured case studies and collaborate with clinicians and interdisciplinary partners to refine protocols and contribute to integrative models of care. This bridges research and real-world application.
Education
We translate complex scientific literature and debates into clinically applicable frameworks and orthomolecular protocols we teach in specific courses, through advanced workshops or in classes and professional training.
Formulation Advisory
We advise producers on nutraceutical and botanical formulations, ensuring ingredient logic, research sourcing and scientific defensibility.
Gastronomic Experimentation from Research Outcomes
Our research extends into applied gastronomy, because it's a way our clients enjoy what they're learning, can get it applicable in their life, and feel that it transcribes into therapeutical action. A structured exploration of cooking methods, raw produce, ecological agriculture, and permaculture systems is in our interest. We study how cultivation practices, soil quality, seasonality, and preparation techniques influence nutrient density, bioavailability, and metabolic impact. We work on continuous recipes that play with nutrients for clinical needs, like inflammation, nootropics, practical dietotherapy and functional foods making. This work connects field to plate integrating food production, sustainability, and culinary transformation into a coherent nutritional framework that is both scientifically informed and practically grounded.
Also delicious!
We support Irish and Spanish local businesses
© 2026 Created with love by Kalavik. Integrative Nutrition & Clinical Research.