Aura: scientific decision intelligence for AI/ML architectures
AURA
SCIENTIFIC KNOWLEDGE LAYER
Evidence-backed decision intelligence
Aura is an evidence-backed decision intelligence system for AI/ML architectures. It transforms literature, benchmarks and experimental results into verifiable technical decisions, tailored to the client’s constraints. Aura not only tells you which architecture to choose, but also what evidence supports that choice, when that choice changes, and what experiments to carry out when science alone is not enough.
Aura starts with the choice the team must make: alternatives, technical constraints, objectives, risks and expected outcome. It does not answer an abstract question: it structures a concrete decision.
Literature, benchmarks and experimental results are related to the alternatives under consideration. Aura distinguishes between what is genuinely comparable and what applies only under different conditions.
Aura shows when the choice changes: latency, budget, privacy, hardware, data quality and operational constraints. The answer is not universal, but specific to your situation.
Aura distinguishes between what the evidence supports, what is missing from the body of evidence, and what science has not yet resolved. No gap is turned into a certainty.
When the evidence is insufficient, Aura proposes the minimal experiment capable of distinguishing between the alternatives. The decision and its outcome inform subsequent assessments.
Every piece of evidence is linked to its source and the reasoning behind it. You can trace the argument back from the conclusion to the source, check the context and distinguish between what is documented and what remains uncertain.
The strength of a conclusion depends on the quality, consistency and applicability of the evidence to your case. Aura makes these factors visible and distinguishes between scientific support and the model’s confidence level.
Considered alternatives, constraints, evidence, recommendations and human judgement are all recorded within the same decision-making process. The Charter can be reviewed, exported and updated without losing sight of the rationale behind the choices made.
Aura does not merely summarise the literature: it applies it to the clients’ constraints. When the sources do not allow for a choice, it transforms the uncertainty into a discriminating experiment and uses the outcome to make subsequent decisions more informed.
Decision · Human-in-the-loop
Aura applies scientific evidence and experimental results to the client’s specific decision and organises them into a verifiable document. Alternatives, conditions, risks and uncertainties remain visible: Aura builds the framework for decision-making, whilst judgement and responsibility remain human. The decision can thus be discussed, justified and updated when constraints change or new findings emerge.
Transparency · Every conclusion can be verified
Aura does not favour any particular technology, model or supplier. Each conclusion sets out the evidence supporting it, the evidence contradicting it, the conditions under which it remains valid, and what cannot yet be determined. It does not ask you to accept an answer: it puts you in a position to verify it, challenge it and make an informed decision.
Connection · MCP Standard
Aura integrates into the workflow where architectural decisions are formulated and transformed into code. Connect its MCP endpoint to the coder or IDE you already use: your assistant will be able to query Aura, submit the use case to it and retrieve the Decision Card without changing your working environment. No servers, models or databases to install on your machine
Sign in to the Aura portal and generate your personal key. From here, you can manage access, Usage and revocation of the connection
Add Aura’s MCP endpoint to your coder or IDE. When you first connect, you authorise the service with your private key; subsequent renewals are handled automatically
Your assistant can now send explicitly selected questions, alternatives and constraints to Aura, monitor the asynchronous processing and retrieve the Card when it is ready
A single HTTPS endpoint. Your coder sends requests and receives responses via the standard MCP protocol — Aura is a remote service that your IDE calls upon as and when needed. It doesn’t matter which IDE you use, provided it supports remote MCP servers with OAuth authorisation.
No packages to install. No local server. No GPU or database on your machine. Aura is a remote service accessed on demand by your coder: all the heavy lifting happens on our end, whilst your IDE remains lightweight.
Output
Aura’s output is not an answer. It is a structured dossier: a reasoned thesis, its citations, counter-arguments, known gaps, a calibrated confidence score and the constraints to be decided. A generic LLM gives you the model’s opinion; Aura provides you with a defensible, reusable, exportable dossier.
Thesis + evidence map for claims + documented counter-arguments + explicit gaps + design constraints subject to human review. To transform a question into a defensible decision.
Typical artefacts: Architectural Decision Record · Verification report · Exploration brief
Research axes classified by status: open, saturated, converging, structural opportunity. Cross-axis tensions, signs of convergence, top bridging opportunities.
Typical Frontier artefacts: 1-pager · State-of-the-art draft · Topic deep-dive
The same payload is both machine-readable (typed JSON fields for downstream pipelines) and human-readable (a human_report to be shown to the user). It can be inspected, exported and cited in a design document or a peer review.
Security · Privacy by design
Aura is designed to apply scientific knowledge to your use case without taking over your development environment. It shares only what you choose to include in the request, retains only the minimum necessary to complete the task, and keeps data, credentials and results separate for each client
Aura only receives the content that you or your assistant explicitly include in the request: the question, the alternatives, the constraints and any necessary context. It does not independently explore repositories, files, branches, history or the output of other tools
Each request is associated with the client that authorised it and is verified at every step. Data, quotas and results remain separate from those of other tenants; no client can query, retrieve or use content belonging to another
Aura receives only the query, alternatives, constraints and context explicitly included in the request. Repositories, files, secrets and output from other tools remain local.
Requests and results are retained only for the stated period and for as long as is necessary for the service. Retention, export and deletion are documented and verifiable.
Identities, credentials, quotas, requests and results are associated with a single scope. Each operation checks authorisation before accessing data.
The clients’ queries, context and results are not used to train shared models. Any signals for improvement are separated from the content and governed by explicit rules.