# Trading Desks to Clinical Trials: Parallels in Applied Vertical AI — Ayush Bhardwaj, Allos AI

## Executive summary

The talk outlines a structured methodology for building and iterating in applied vertical AI, arguing that model infrastructure and APIs are commodities. The true 'moat' lies not in the technology itself, but in proprietary data and deep domain expertise. The process requires narrowing the problem scope, curating unique datasets (like failed experiments or trade theses), and crucially, integrating a human domain expert ('the user') to create an endless learning loop that guides prompt refinement and error analysis.

## Key takeaways

- The Moat is Domain Expertise: In vertical AI, the competitive advantage does not come from the model (e.g., Claude or ChatGPT) or the infrastructure; it comes from proprietary data and domain expertise which must be curated by human experts.
- Focus on Narrow Tasks: When formulating a problem for an AI agent, do not try to solve everything at once. Break down complex tasks into very narrow, specific steps (e.g., ranking stocks based on IT capital expenditure in US equities).
- The Importance of Proprietary Data: Generic data sources are insufficient. The most valuable assets are proprietary datasets that are expensive to acquire or withheld by industry players (e.g., failed clinical trials, internal trade theses).
- Hire the User, Not Just the Model: The single most critical step is hiring a domain expert ('the user'). This person guides the tool by curating sources, refining prompts, and performing judgment—a process that transforms generic AI tools into industry-specific assets.

## Technical details

- Applied Vertical AI Definition: Applied AI built for one very specific industry (e.g., building drugs with AI at a pharma tech company, rather than general translation).
- AI Agent Production Readiness: The key question is not if agents are deployed, but whether they justify their ROI and save or make money end-to-end.
- Model Iteration Techniques: Methods for improving models include Supervised Fine-Tuning (SFT), Reinforcement Learning from Human Feedback (RLHF), and Error Analysis (the highest ROI method, involving analyzing logs to correct model errors without touching weights).
- Causation vs. Correlation: For models to make good decisions in complex fields like finance or pharma, they must move beyond pattern matching (correlation) and achieve true understanding of cause-and-effect (causation).

## Practical implications

- Build engineers should focus on building robust data pipelines that ingest and structure proprietary, unstructured organizational data.
- Design AI agents with narrow scopes and modular steps rather than attempting single, complex tasks.
- Implement strong observability tools to track model behavior and facilitate error analysis, which is the most cost-effective way to iterate.
- Structure workflows around human domain experts who act as prompt engineers and quality judges (AI in the loop - AITL).

## Topics

Artificial Intelligence, Machine Learning, FinTech, PharmaTech, Prompt Engineering, Build Systems, Stanford AI Index report, JP Morgan / Morgan Stanley reports

Source: https://www.youtube.com/watch?v=Yphdry8ttAQ