androidengineers.Book a session

Retrieval agents and research workflows

Build an evidence-seeking research workflow

articleSelf-paced

What you will learn

Query rewriting, Evidence ledgers, Contradictions, Citations, Search budgets.

Engineering the capability

A research agent needs a stopping rule tied to evidence, not a feeling that the answer is complete. Break the question into claims, record which sources support each claim, and identify unresolved contradictions. A source count is not the same as independent evidence: several pages may repeat the same underlying statement.

Query rewriting can improve recall but can also drift from the user’s question. Preserve the original task and record why each new query is needed. Bound search depth and total calls. If the remaining evidence is insufficient, return a partial answer and describe the gap.

Maintain an evidence ledger with source identity, relevant passage, retrieval time, and supported claim. Keep content permissions and source freshness in the retrieval layer. The final writer should use that ledger instead of inventing references from memory.

Worked case

Two policies disagree on a cancellation deadline. The agent should compare versions and applicability rather than choose the first result. If neither has a trustworthy effective date, the correct output may be a conflict requiring review. Extra searches are useful only if they can resolve that specific uncertainty.

Put it into practice

Continue with the next lab: research with a bounded evidence ledger. Build the artifact, record the failure cases, and explain the tradeoff before moving on.

YOUR LEARNING JOURNEY

0 of 118 available lessons completed

Progress saved in this browser. No account needed.
Build an evidence-seeking research workflow | Agentic AI | Android Engineers