Mike Conover cofounded Brightwave, where he served as chief executive while developing AI systems for investment research, and helped create Databricks Dolly, an early instruction-following language model built with openly available technology. His work addresses a central problem in financial AI: how automated systems can analyze vast document collections without obscuring the evidence, uncertainty, and judgment behind their conclusions.
From network science to financial AI
Conover earned a doctorate in complexity science at Indiana University and contributed to Truthy, a project studying information diffusion across social networks. His research on political communication found pronounced ideological separation in retweet networks even when direct exchanges crossed partisan boundaries. He also helped develop methods for visualizing large communication networks.
He subsequently worked at LinkedIn, including on homepage-news relevance, and became director of financials machine learning at Workday. At Databricks, he led open-source language-model engineering and coauthored the March 2023 introduction of Dolly and its instruction-tuning approach. The subsequent Dolly 2.0 release included an openly available dataset of 15,000 human-generated instruction-and-response examples, expanding the practical foundations for commercially usable open models.
Conover cofounded Brightwave with Brandon Kotara and was its chief executive when the company announced a $6 million seed round in June 2024. Brightwave built research agents to examine filings, earnings transcripts, vendor contracts, and diligence materials; a $15 million Series A followed later that year. Conover has highlighted changes in earnings-call guidance and litigation disclosures as consequential signals these systems should surface.
Independent factual verification: Small extraction errors compound across multistep workflows; separate verification calls should test whether claims follow from their sources and preserve relevant dates and context.
Focused synthesis across documents: Large context windows do not guarantee faithful analysis. Decomposing investigations into specific questions produces denser findings and makes cross-document reasoning easier to inspect.
The latency trap: Slow research agents limit how often analysts can refine instructions, making response time essential to both usability and the quality of human oversight.
Interrogable findings: Analysts should inspect supporting documents, challenge individual passages, and redirect investigations toward emerging risks or unexplored evidence.
Conover favors small, composable tools over elaborate casts of anthropomorphized agents. His objective is research infrastructure that strengthens professional judgment while keeping consequential conclusions open to examination.