FDFeiyu Deng
CV中

AI-assisted research systems

Tools for making complex research inspectable

I build workflows for literature analysis, symbolic derivation, numerical verification and manuscript review. The examples below describe the problems, methods and outputs.

01

Derivation and verification

Problem
Long calculations hide conventions, sign choices and intermediate identities.
Workflow
Equation extraction, dependency mapping, symbolic reconstruction, limiting-case tests and mismatch logging.
Human judgment
The researcher decides which assumptions are legitimate and whether agreement is physically meaningful.
Output
A reproducible derivation note with assumptions, verified steps and unresolved points separated.
Quality checks
  • Check original sources
  • Derived and assumed claims remain distinct
  • Independent checks where possible
02

Literature and manuscript intelligence

Problem
Claims, notation, citations and reviewer requests drift across manuscript versions.
Workflow
Claim-to-source tracing, notation checks, version comparison, bibliography validation and overclaim review.
Human judgment
Review each scientific claim, citation and response against the source material.
Output
An audit record that links each finding to a source location and a review status.
Quality checks
  • Source spans and version hashes
  • Track open questions
  • Use published material for public examples
03

Computational and visual prototyping

Problem
A continuum calculation needs a numerical model, convergence tests and figures with clearly stated regimes of validity.
Workflow
Model specification, discretization, convergence tests, continuum comparison and figure generation.
Human judgment
The model, regulator and observable are fixed before visual polish begins.
Output
A self-contained Mathematica or Python record with explicit errors and reusable figures.
Quality checks
  • Convergence before interpretation
  • Record data sources and generation methods
  • Figures carry provenance and alt text