All projects View code
Trustworthy AI Research
Deep Research Agent with Citation Graph
A multi-hop research agent that expands questions, retrieves evidence, builds a citation graph, grades sources, detects contradictions, and produces a traceable report.
THE PROBLEM
Why this exists.
Research agents can sound convincing while hiding weak evidence, conflicting sources, or unclear provenance.
THE APPROACH
How I approached it.
The system models sources as a directed citation graph, grades source quality, links contradictions, and compiles findings with inline citations so conclusions remain inspectable.
ARCHITECTURE
The system flow.
Research query
Sub-question expansion
Retrieval
Hop chain
Citation graph
Source grading
Contradiction detection
Structured report
ENGINEERING SIGNALS
What to notice.
- Up to 10 research hops
- Directed citation graph
- A–F source grading
- Contradiction detection
- Human-confirmed file writes
WHAT I LEARNED
The takeaway.
Trustworthiness improves when evidence is a first-class data structure instead of an afterthought. Citation provenance, source quality, and disagreement between sources should be represented explicitly.