70–80% of chronic disease burden is attributable to environmental exposures. We built the platform to act on that.
vasus.ai is grounded in exposomics — the scientific discipline studying the totality of environmental exposures across a lifetime and their relationship to health. Every weight, every signal, every recommendation is derived from peer-reviewed literature. This page explains what we built and why.
The exposome: the totality of environmental exposures across a lifetime.
The term exposome was coined by Christopher Wild in 2005 to describe the cumulative environmental exposures a person encounters from conception to death — and their biological responses to those exposures. Wild's insight was that the genome alone cannot explain the burden of chronic disease. The missing variable is the environment.
The science has since been institutionalised at the highest levels of environmental health research. Exposome-NL at Utrecht University, ISGlobal in Barcelona, and the Human Early Life Exposome (HELIX) consortium have collectively advanced the field into a recognised discipline. The weight of evidence is now unambiguous: the environment is not a secondary factor in chronic disease — it is a primary one.
Despite this, no platform existed that connected real-time environmental conditions to individual condition management with peer-reviewed evidence. Air quality apps give numbers. They do not explain what those numbers mean for a person with migraines, asthma, or a cardiovascular condition. vasus.ai closes that gap.
How 350,000+ papers become a cited recommendation.
The corpus is not a search engine. Each query triggers a five-stage retrieval and synthesis pipeline designed to surface the most relevant evidence for a specific condition, location, and environmental context — and to discard irrelevant or low-quality matches before synthesis.
PubMed corpus covering 18 condition categories. Ingested weekly via NCBI Entrez. Embedded using BAAI/bge-base-en-v1.5 (768-dimensional vectors).
PubMed® is a registered trademark of the U.S. National Library of Medicine. vasus.ai is not affiliated with or endorsed by NLM or NIH.
PostgreSQL pgvector extension with HNSW (Hierarchical Navigable Small World) indexing. Approximate nearest-neighbour search at scale with sub-second retrieval.
Query planner → pgvector search → BM25 + semantic ranker → LLM relevance gate → synthesis. Each stage eliminates low-quality matches before passing to the next.
Not every retrieved paper makes it into the response.
After semantic retrieval and ranking, a separate LLM call evaluates each candidate paper against the user's specific condition context. Papers that are topically adjacent but clinically irrelevant — or that describe mechanisms unrelated to the environmental signals present — are filtered out before synthesis. This gate is the primary defence against hallucinated or weak citations.
Grounded in evidence. Explicit about uncertainty.
The synthesis step uses OpenAI gpt-4.1-mini to generate structured output: risk level, a synthesis paragraph connecting the environmental signals to the clinical literature, condition-specific recommendations, and uncertainty notes. The model is explicitly prompted to acknowledge gaps and avoid overclaiming. Every claim is grounded in a retrieved paper.
Seven signal types. Three Google Environmental APIs.
Every tile refresh makes 7 API calls across three Google services. Raw signals are stored, then computed into 24h and 72h window features used by the EHSPI and the insight pipeline. Updated hourly for highest-priority tiles.
Barometric pressure is the strongest replicated environmental trigger for migraine.
The mechanism is well-established: rapid changes in atmospheric pressure — associated with frontal weather systems, Chinook/foehn winds, and altitude change — trigger trigeminal nerve activity in susceptible individuals. The effect is dose-dependent and lag-dependent, with most studies reporting a 6–48 hour window between the pressure event and migraine onset.
vasus.ai tracks both the 24-hour pressure delta (the magnitude of change) and the 72-hour pressure volatility (the standard deviation of pressure over three days). Chronic migraineurs show sensitivity to sustained instability, not only acute drops — the 72h volatility signal captures this.
Secondary signals include heat index, peak temperature, and dew point. Air quality signals (PM2.5, AQI) are included at low weights — the association with migraine is documented but modest relative to the pressure pathway.
PM2.5 and pollen are co-primary drivers of respiratory exacerbation.
Fine particulate matter (PM2.5) is among the most robustly established environmental health hazards. Particles under 2.5 micrometres penetrate the alveoli, triggering inflammatory responses in the airways. The PM2.5-asthma dose-response relationship is linear and well-replicated across dozens of population studies. At vasus.ai, PM2.5 mean and peak are tracked separately — chronic exposure and acute episode risk operate through different mechanisms.
EHSPI v1.6 added pollen as a co-primary signal for respiratory sensitivity. Outdoor pollen is the leading precipitant of asthma exacerbations globally — a fact that is underrepresented in most environmental health tools. The pollen exposure load signal is a recency-weighted composite across tree, grass, and weed pollen.
Ozone peak is tracked separately: ozone operates through a different pathway from PM2.5 (oxidative stress rather than particulate deposition) and is an independent asthma exacerbation driver at concentrations above WHO guideline thresholds.
Sustained heat index is a stronger cardiovascular signal than peak temperature alone.
Cardiovascular risk from heat is driven by sustained thermal load, not individual temperature spikes. The body's heat dissipation mechanisms — sweating, vasodilation, increased cardiac output — become progressively overwhelmed over days of elevated apparent temperature. Heat index integrates temperature and humidity into a single apparent temperature measure, making it a more clinically relevant signal than dry bulb temperature alone.
PM2.5 is a co-primary signal for cardiovascular sensitivity. The PM2.5-cardiovascular mortality relationship — operating through systemic inflammation, endothelial dysfunction, and autonomic nervous system effects — is one of the most replicated findings in environmental epidemiology. The 2019 Global Burden of Disease study identifies PM2.5 as the third leading risk factor for cardiovascular mortality globally.
The overnight hot hours signal (hours above 24°C between 22:00 and 06:00) captures a distinct mechanism: thermal disruption of sleep → sleep deprivation → cardiovascular strain. This is modelled separately from the daytime heat burden.
Overnight temperature above 24°C is the primary environmental sleep disruptor.
The thermal environment of the sleeping space is the dominant environmental determinant of sleep quality. The human thermoregulatory system during sleep is highly sensitive to ambient temperature: core body temperature must drop by approximately 1°C to initiate and maintain sleep. When ambient temperature prevents this drop — particularly in humid heat where evaporative cooling is impaired — sleep architecture is disrupted.
Obradovich et al. (2017) analysed 765,000 US survey responses and found that a 1°C increase in monthly minimum temperature is associated with 0.58% more insufficient sleep nights per person. At population scale, this is a significant health burden. vasus.ai tracks hours above 24°C specifically between 22:00 and 06:00 — the overnight window that directly affects sleep.
Dew point above 21°C is tracked as a secondary signal — this threshold represents oppressive humidity that independently disrupts sleep even at moderate temperatures. Pressure volatility is included at moderate weight: barometric changes affect sleep architecture in sensitive populations independent of temperature.
Pollen exposure load carries the highest single weight in EHSPI v1.6 — 0.40.
Outdoor pollen is the leading cause of allergic rhinitis globally, affecting approximately 400 million people. Unlike PM2.5 — which affects all populations at sufficient doses — pollen sensitisation is individual-specific: a person sensitised to birch tree pollen may have minimal reaction to grass or ragweed, and vice versa. The three pollen types (tree, grass, weed) are tracked separately via Google's Universal Pollen Index (UPI), then combined into a recency-weighted composite.
The pollen-PM2.5 compound effect is an increasingly documented phenomenon: fine particulate matter carries pollen sub-particles and pollen fragments, amplifying the allergenicity of a given pollen load. This synergy means that high-PM2.5 days during pollen season are disproportionately more harmful than either signal in isolation.
Ozone increases nasal epithelial permeability, amplifying sensitisation to pollen. Dew point affects pollen dispersal and rupture — high humidity causes pollen grains to burst, releasing smaller sub-particles that penetrate deeper into the airway.
Evidence-grounded. Transparent. Live.
Every weight published. Every citation traceable. Try the platform or engage with the research team.
Select a sensitivity and a city. See exactly what the evidence-based intelligence output looks like in practice. No account, no data stored.
Open the app →All five sensitivity weight tables, 20-city coverage, live EHSPI scores, confidence levels, and the Google API validation study — published transparently.
View research →Discuss methodology, weight validation, or data collaboration. We are actively seeking research institution partnerships for formal EHSPI validation.
Contact research team →