AI, Trust, and the Future of Financial Transparency with David Trainer of New Constructs
How do you trust the numbers? In this episode, David Trainer breaks down how New Constructs combines machine learning, NLP, and deep accounting expertise to reverse accounting distortions and surface the true economics of a business. We dig into their RoboAnalyst system, why ontology + taxonomy matter more than flashy models, and how a narrow, expert-defined data domain—what David calls a “small language model”—can power auditable, repeatable alpha. We also talk transparency (no black boxes), an ETF and indices that have outperformed the S&P 500, and New Constructs’ Google Cloud partnership to build an AI investing agent. What you’ll learn: ● Why “garbage in, garbage out” still rules—and why data quality and footnotes are the edge ● How New Constructs structures filings (MD&A, footnotes, corporate actions) into an auditable knowledge graph ● The human-in-the-loop design: rules engines + auto-parsing for the easy stuff, experts on the exceptions ● Why transparency and auditability (not model secrecy) build institutional trust ● The case for domain-specific small language models vs “internet-sized” LLMs ● Thoughts on ESG (start with G—governance/ROIC), regulatory change, and error-proofing at scale ● Where this goes next: AI agents for fundamentals, technicals, sentiment, and potential expansions to global coverage and even political/healthcare integrity scoring Chapters: 0:00 Intro & origin story (the “apples-to-apples” accounting problem) 2:40 Mission: improving the integrity of capital markets 5:14 Reversing accounting distortions (economics vs. accounting) 7:30 Inside RoboAnalyst: rules, NLP, and human validation 11:07 What differentiates New Constructs (methodology, transparency, alpha) 12:24 Why data plumbing beats model hype 13:32 “Small language models” for trusted finance AI 15:40 Training signals from MD&A & footnotes; seeding the rules 19:52 Data checks, model sanity, and exception handling 21:13 Automation vs. analysts: serving humans only the hard problems 25:18 Trade-offs (scale, capital) and the Google Cloud partnership 29:31 Building trust: 100% auditability, no black boxes 33:49 Guardrails against errors & overconfidence 36:41 Adoption challenges & Wall Street incentives (IPOs, “adjusted EBITDA”) 41:17 Accounting rule changes & disclosure shifts 42:47 Roadmap: agents, broader data, making markets boring (and better) again 47:05 What could threaten the edge; why the source dataset is the crown jewel 49:22 ESG, governance, and ROIC as the signal that matters 54:04 Moonshots & where this goes in 3–5 years 56:05 Wrap-up
