AI summary by TwinMind
Human Stories in AI: Brian Risk@devra.ai
StatQuest with Josh Starmer
35m 48s
Career Journey and Background
- Brian Risk: multi-talented data scientist, president/founder of Devra (automated coding company), musician
- Age 48, single father of 3 kids, maintains full-time job while running company
- Diverse career path: bartender at music club, high school math teacher, entry-level programmer
- Math degree background, transitioned into data science as field emerged
- Always had affinity for data throughout various career transitions
Bioinformatics Experience
- Landed bioinformatics job analyzing protein data from tumors and DNA data
- Core research: proteins found in cancers potentially originating from non-coding DNA regions
- 99.5% of genome not supposed to code for proteins - when it does, problems arise
- Developed software with specific criteria:
- Match large amounts of protein data to genomic data
- Required extreme speed and accuracy to avoid false positives
- Published 2 papers during this period
- Lab unfortunately imploded and dissolved, great work never moved forward
Transition to Advertising Data Science
- Financial necessity with kids led to advertising industry ("dark side")
- Data science in advertising: reduce ad costs by targeting only interested people
- Predictive modeling using:
- Time of day user sees website
- Browser type
- Geographic characteristics (state)
- Referring website (important factor)
- Recent example: English muffins analysis revealed car enthusiasts don't like English muffins
- Used Fisher's exact test, sorted by p-values
- Lowest p-values for negative indicators were all car-related websites
Core Philosophy: Always Be Curious (ABC)
- References StatQuest "ABCs of Data Science" video
- Curiosity about data = fundamental key to success
- Granular level understanding: individual records vs. billion-row databases
- Need curiosity about results produced - self-validation essential
- People rely on data scientist outputs for decision-making
- Must try different methods (Random Forest vs. K-means) and validate approaches
AI and Coding Automation
- AI explosion in 2022 with OpenAI APIs enabling custom applications
- Common concerns: fear of replacement, criticism of bland/unoriginal output
- Coding benefits from "bland" - want standardized, recognizable patterns
- 80% of coding involves previously solved problems (login screens, K-means analysis)
- Developers constantly "reinventing the wheel" - documented solutions exist on Stack Overflow
Devra Development and Functionality
- Born from idea to eliminate copy-pasting when using AI for coding assistance
- Current workflow: copy function, paste supporting classes, type question to ChatGPT
- Devra allows AI to explore existing codebase autonomously
- Simple prompts like "write unit test for this function" - AI finds function, analyzes content, adds to test suite
- Current version: hundreds of thousands of lines of code, 90% AI-generated, 10% human refinement
AI Capabilities and Limitations
- AI gets you 80% of the way there
- Remaining 20% requires true problem understanding and coding skill
- Metaphor: shoveling vs. excavator operation
- Everyone can shovel, but excavator driver needs more skill/certification
- Much more productive but requires higher expertise
- AI won't replace jobs but will shift skill requirements
- Advice: work with AI intensively now, learn tools, don't give up
User Experience and Workflow
- Zero-to-one capability: start projects from scratch with CSV loading, analysis specifications
- Can enhance existing projects
- Conversational interface: iterative back-and-forth refinement
- Example: creating JavaScript game with 11-year-old son Max
- "Give it vintage look" resulted in TV screen appearance with scan lines
- 2-minute turnaround for initial results
- Follow-up requests: responsive design, color scheme changes, modern styling
Business and Entrepreneurship Advice
- Started business at 48 with kids and full-time job
- Two ABCs: Always Be Curious + Always Be Creating
- Create what you're passionate about vs. interviewing 100 potential customers
- Limited time (9 PM after kids' bedtime, 5 AM before family wakes)
- Work on what you have energy and drive to complete
- Steve Jobs quote: "Real artists ship" - difference between hobby and viable product
- Prioritize minimal viable features, stick with it, show to friends
- Many previous unsuccessful attempts, but Devra gets immediate understanding from users
Product Validation
- First product where everyone immediately understands the value proposition
- Eliminates template searching - provides immediate first-order approximation
- Handles file creation (CSS, HTML, JavaScript) with functional placeholder content
- Gets users 70-80% to goal, leaves fun customization work to humans
- Reduces multi-day setup to immediate start with refinement focus
