Not one general-purpose assistant answering questions about stocks — a set of
specialised agents, each investigating one domain and reporting what the evidence supports.
Why several agents rather than one
A single model asked "is this a good company?" produces a fluent answer with no
way to check which part came from a filing and which came from its own prior. Splitting the work
by domain makes each claim traceable: the dividend agent reads dividend lines, the balance sheet
agent reads the balance sheet, and each reports its own confidence and the period it looked at.
What each agent returns
Metrics — with source, reporting period, currency and unit attached to every figure.
Signals, risks and opportunities — stated separately, so a risk is never buried inside a positive summary.
Confidence and data quality — how much of the analysis rests on complete data.
Reasoning — why the agent reached its rating, in plain language.
What the agents will not do
Invent a figure that is not in the filings. Missing data is reported as missing.
Mix quarterly and annual periods, or adjusted and unadjusted figures, without labelling them.
Produce a headline research score when coverage is below the threshold — the score is withheld and the coverage shown instead.
Present a scenario as a forecast. Projections are labelled AI-generated assumptions.
Angaza Hisa is an analytical and educational tool, not investment advice. Market data is sourced from official exchanges and feeds and may be delayed or incomplete. AI scores are analytical signals generated from available market data and are not investment recommendations.