Reading a factor research report is rarely the hard part. Implementing it is.
Before a factor can be validated, someone has to translate formulas into code, map variables to database fields, debug execution errors, and run evaluation with tools like Alphalens. If the results aren't satisfactory, the entire process starts over.Much of factor research is still repetitive engineering rather than research itself.
The Research Report Analysis & Factor Derivation module in DolphinDB Starfish automates that entire pipeline—from report parsing to factor evaluation.
One PDF Away from a Runnable Factor
Turning a research paper into a production-ready factor traditionally involves three major hurdles:
Writing implementation code
Fixing execution errors
Running factor evaluation
With Starfish, the workflow becomes three simple steps.
Step 1. Upload a Research Report
Simply upload a PDF. AI automatically extracts the factor name, mathematical formula, variables, and economic intuition from the report. What used to require manually reading and annotating the document can now be completed in minutes.
Users can still review and edit the extracted content when necessary, combining AI automation with human verification.
Step 2. Choose Your Data Sources
Select the database tables AI is allowed to access, and let it handle the rest.
Step 3. Let AI Build, Test, and Evaluate the Factor
This is where most of the time savings come from.
After you select the data tables, Starfish automatically understands the schema, maps report variables to database fields, generates factor code, and executes it in a real environment. If the code fails, the error is fed back to the language model, which fixes the implementation and retries automatically until it runs successfully. Once it does, Starfish immediately evaluates the factor with Alphalens.
AI Doesn't Just Write Code—it Fixes Itself
One question naturally comes up: Can AI-generated code really be trusted?
The reality is that execution failures are expected. The important part is what happens next.
Every implementation generated by Starfish is executed immediately. If it fails, the runtime error is fed back to the language model, which analyzes the problem, regenerates the code, and retries automatically until the implementation executes successfully.
The process goes beyond code generation. If the model determines to adjust the factor definition itself, it also updates the formula and its documentation so that the implementation and the explanation remain consistent.
The result is a self-correcting workflow that spans the entire factor development process—from extracting formulas and generating code to debugging and evaluation. Researchers only need to answer one question at the end: Is this factor worth using?
Automatically Generate Better Variants
One of the most practical capabilities in Starfish is automatic factor derivation.
Researchers typically spend hours tweaking parameters, rerunning evaluations, and comparing results.
Starfish turns that into an automated loop. Set the number of iterations, and the AI generates new factor variants, evaluates each one, and keeps improving them until every round is complete. It then compares all candidates using metrics such as the Information Coefficient (IC) and recommends the strongest performer with visual reports, so researchers can focus on evaluating ideas instead of comparing results.
Make Existing Knowledge Work for You
For teams with an existing library of factor implementations, Starfish also supports Retrieval-Augmented Generation (RAG).
When generating factor code or fixing execution errors, Starfish first retrieves similar implementations, past debugging records, and other relevant knowledge from the team's knowledge base, providing the language model with the context it needs to generate code that follows the team's best practices.
As the knowledge base grows, so does the quality of the AI's output. Instead of remaining locked away in documents and repositories, a team's accumulated expertise becomes part of every new factor development task.
Enterprise-Ready Permission Control
For enterprise deployment, Starfish provides fine-grained AI governance.
Administrators can configure:
Task quotas to limit how many report-analysis or factor-derivation jobs each user can create, helping control AI usage costs.
Data access permissions at the table level, ensuring AI only accesses explicitly authorized datasets.
Usage monitoring to track AI consumption across users and teams.
After deployment, users can also choose from multiple supported large language models within the AI Chat interface, ensuring the quality of AI-generated output meets the team's standards.
Rethinking the Entire Factor Research Workflow
Building a factor is no longer the bottleneck. Researchers can upload multiple reports, let AI implement and evaluate them in parallel, and focus on what humans do best: deciding which ideas are worth pursuing.
From report parsing to optimization, DolphinDB Starfish automates the repetitive steps of factor development, making the entire workflow faster, more scalable, and easier to iterate.
Download the Starfish now.
Thanks for reading! Stay connected with DolphinDB by following us on X and LinkedIn for the latest updates, technical articles, and product news. For more information, visit our website.
Top comments (0)