AI at Investboard
Investboard uses AI in six clearly bounded places: for language, summaries and context. Not for calculations, forecasts, orders or decisions.
The language model enters only after Investboard has calculated, checked and limited the data to the purpose at hand. Every AI function follows the same principle: the evidence stays visible and the decision stays with you.
We count every independent function path backed by Bedrock. A use may explain, summarize or match an existing template. None may control your wealth.
Sequence matters
Calculate first. Phrase second.
Rules and calculations establish the finding. AI makes it readable.
The model has no open access to Investboard. Every request follows a bounded path.
Calculations, rules and thresholds are produced without a language model.
Only fields intended for the particular purpose are assembled.
We do not rely on a label. The Bedrock integration sets concrete boundaries for region, access and logging.
Every model call uses a geographically bound EU profile in AWS Bedrock. The shared client technically rejects any model identifier without the “eu.” prefix.
Investboard connects to AWS Bedrock in eu-central-1. The EU profile may distribute processing within the European AWS geography, not outside that geography.
Requests travel over the encrypted AWS interface. Access uses role-based AWS permissions under the principle of least privilege.
Investboard does not enable optional Bedrock model-invocation logging to CloudWatch or S3 in its AWS infrastructure.
AWS states that Amazon Bedrock inputs and outputs are not used to improve base models and are not made available to model providers.
For every purpose, the application assembles a fixed data shape. Raw database rows and technical access rights are not passed through to the model.
Language models can be wrong. Investboard therefore never treats an answer as the sole source of truth.
Overlong, off-topic or manipulative inputs are bounded. Common personal-data patterns trigger a visible warning.
Figures in outputs are checked against the supplied context or a fixed list of permitted values.
Buy and sell instructions, guarantees and certain claims about the future are excluded. The user decides.
Severe violations are blocked. Where fixed wording is available, the surface falls back to it.
Blocking decisions and many interactive calls are logged for technical review. Blocked quarterly letters are recorded as blocked and are not delivered.
For Amazon Bedrock properties, we link to current AWS documentation. Investboard-specific boundaries are additionally enforced in code and tests.
No. Investboard uses AI to explain calculated findings, summarize content and match existing templates. The model cannot place an order, move money or make a decision for you.
Requests start with Amazon Bedrock in Frankfurt and use a geographically bound EU inference profile. AWS may distribute processing within the European AWS geography, but cannot route it to another geography.
See Investboard in the demo account or read how the rest of the security architecture protects your data.
The central client accepts only an EU inference profile and starts in Frankfurt.
Inputs, numeric grounding, advice boundaries and output shape are checked.
Only a passing answer appears. Otherwise it is blocked or replaced by fixed wording.
Free text remains free text: if you type personal data into a question, Investboard detects common patterns and warns you. The text is not silently changed. Do not submit such details when they are not needed for your question.
Under Amazon Bedrock's published terms, customer inputs and model outputs are not used to improve base models and are not made available to model providers. Investboard does not use them to train its own models either.
Yes. Language models can make mistakes despite guardrails. Investboard therefore precomputes figures, checks outputs against context, blocks severe violations and deliberately keeps decisions outside the model.