An agentic research engine that turns a plain-language question into fact-checked, source-cited writing.
Generative writing tools are fluent and unaccountable. They produce a paragraph and leave you to work out which half of it is true. For research work that is worse than useless - the cost of checking the output exceeds the cost of writing it yourself. The question was whether verification could be built into the pipeline rather than bolted on after it.
Six agents in a fixed sequence, each with one job. A query optimiser expands a plain question into a richer search brief; a research agent gathers from multiple sources; a claim extractor pulls out the assertions that actually need checking; a verifier cross-checks those claims against trusted sources; a report generator produces the fact-check artefact; a drafting agent writes the final piece in the requested style. Claim extraction and verification are their own stages, not a filter at the end - that is the architectural bet the whole system rests on.
A complete product with frontend, backend and AI layers. Output is split across four tabs - research, fact-check report, final draft, references - so a reader can walk the process rather than trusting the result. The templating layer formats the same verified research into whichever content style the user picked.