01
The problem
Research needs a reviewer, not just a generator.
Paper reading is not only a summarization task. Researchers need to know which claims are supported, how studies connect to a question, and where the evidence is still weak.
2026 · Personal
Sorot
An AI research-review assistant for reading papers with the discipline of a peer reviewer.
Sorot helps researchers inspect claims, trace evidence, connect studies, and build a working synthesis without treating AI-generated text as the final authority.
02
The solution
Review every claim like a peer reviewer.
Sorot turns AI into a structured research companion that helps inspect evidence before it becomes part of a synthesis.
- Upload and organize the papers relevant to a question.
- Inspect claims against the source text and evidence.
- Connect studies and select the papers that belong in the synthesis.
03
The product
A working loop from paper to synthesis.
The finished application supports a deliberate workflow: start with a research question, work through the literature, check the evidence, and export a synthesis that remains reviewable.
04
The principle
AI assists. The researcher decides.
Sorot is designed around human review. AI suggestions are useful for speed, but acceptance, rejection, and the final interpretation stay with the person doing the research.
05
The builder
One person, from idea to finished product.
I created Sorot end to end: the product idea, research workflow, interaction model, implementation, and public launch.
Problem
Researchers in industry and academia need to move through large bodies of literature, but ordinary AI tools tend to summarize before they verify. That makes it easy to lose the source, overstate a finding, or mistake a plausible draft for a defensible review.
Approach
Created the product end to end—from the idea and research workflow to the finished application. Sorot treats AI as a review assistant: it helps inspect claims, locate supporting passages, compare papers, and assemble synthesis while keeping the human researcher responsible for the final judgment.
Outcome
Finished and publicly available as a focused research workflow for turning papers into evidence-aware review work.