All work

AI in Venture Capital

Does AI make venture investors better, or just more confident?

Undergraduate researchAI in Venture CapitalDoes AI make investors better, or just more confident? 41 papers screened for a 5-paper pilotWith Prof. Solmaz Batebi, 2026–present
Type
Undergraduate research, University of Washington Bothell
Guidance
Professor Solmaz Batebi
When
October 2026–present
Status
Phase 1 complete

The question

Venture firms increasingly use AI to screen startups. The research on whether it actually improves their decisions is spread across finance, management and computer science journals. Before I can propose an original research question, I need to know what has been studied and where the gaps are.

My approach

Rather than running searches and reading abstracts by hand, I built a pipeline that does discovery and screening systematically. I review every conclusion myself; the pipeline only narrows the field.

Research pipeline. Phase 1, complete: research question, AI-generated search queries, search of three databases (46 hits), deduplication and relevance ranking (41 unique), pilot selection (5 papers). Phase 2, next: retrieve full texts, analyze each paper, compare across papers, research-gap report, review with Prof. Batebi. Phase 1: complete OpenAlex, Semantic Scholar, SSRN Phase 2: next Researchquestion AI-generatedsearch queries Search 3databases Deduplicate,rank relevance Select pilotpapers 46 hits 41 unique 5 selected Retrievefull texts Analyze eachpaper Compareacross papers Research-gapreport Review withProf. Batebi
Green steps are complete; dashed steps are the next phase.

Results so far

Phase 1 is complete. The first live run returned 46 papers from 2020 onward. After merging 5 duplicates, 41 were unique, and I selected 5 for an initial pilot.

Two directions look promising. These are open questions, not findings:

  • Does AI help investors make better early-stage decisions, or mainly make them more confident in automated recommendations?
  • Does explainable AI improve decisions for highly novel startups, where past data says the least?

Next

Retrieve full texts, expand to 20–50 papers, compare methods and limitations across them, and write one research-gap report to review with Professor Batebi.

Technical details

Python; Gemini API for query generation; OpenAlex, Semantic Scholar and SSRN for discovery; SQLite and JSON caching. The pipeline retries failed API calls and saves partial results rather than treating them as complete. 25 automated tests.