AI summary by TwinMind
Human Stories in AI: Khushi Jain
StatQuest with Josh Starmer
27m 12s
Personal Background and Early Journey
- Kushi Jane: Data analytics development at John Deere, pursuing Master's in Computer Science Data Science at University of Illinois Urbana-Champaign
- Never intended to code from middle school, considered herself "terrible at tech"
- Science Olympiad geek, loved all kinds of sciences
- Junior year high school: Dad suggested computer science class
- Discovered "crazy rush" when code worked, instant gratification feeling
- Didn't get into computer science at UIUC, took Information Science instead
- Information Science provided balance: technical side + human-centric approach
- Customer orientation, data presentation, technology presentation
- Appeals to different audiences
Illinois Data Science Club
- Co-founded with roommate Ria Sha (same major at UIUC)
- Goal: Create low-pressure community vs hardcore hackathons/tech consulting
- Safe space for different experience levels to learn data science
- Project-based learning: students choose own projects, receive guidance
- Showcase presentations at end
- Core motto: "We don't just do data for data, we do it to solve bigger problems (business or societal)"
Academic Path and Current Status
- Senior graduating, planning Master's in Computer Science Data Science (MCSDS) at UIUC
- Online program, self-paced (1-3 courses per semester)
- 8-course program, transferring 2 undergrad courses
- Plan: Complete in ~1 year vs typical 2 years
- Working full-time while pursuing degree
- Minor in Computer Science, took diverse courses including computer architecture
John Deere Internships
First Internship (End of Sophomore Year)
- Customer call sentiment analysis project
- Grouped calls into red/yellow/green categories based on sentiment
- Used basic Python, scikit-learn packages, feature engineering
- Key learning: Data quality matters most
- Challenges:
- Poor audio transcription quality
- Humans couldn't classify calls clearly
- Many calls weren't actual customer calls (John Deere field workers)
- Different tones/contexts for different speaker types
- Lesson: "ML is not magic - if humans can't do it well, ML models likely won't either"
- Used TF-IDF (Term Frequency-Inverse Document Frequency) for NLP analysis
Second Internship (Post-ChatGPT Era)
- Factory Automation focus
- Problem: Improve completeness/accuracy of machine defect documentation
- Challenge: Operators must identify specific parts from hundreds of thousands of options
- Solution: Semi-automated approach combining AI + human intelligence
- Technical approach:
- Used bill of materials in vector database setup
- Hierarchical BOM structure (broad → specific machine parts)
- Embeddings-based similarity searches
- Series of narrowing questions (top 5 parts → user picks → deeper level)
- 200,000 parts total, categorized approach
- Philosophy: Don't rely completely on AI, merge with human intelligence for best output
Disruption Lab Experience
- Tech consulting academic unit within Gies College of Business
- Student groups work as software engineers for startups/small companies
- Real-world project experience vs classroom learning
- Created live crypto search engine project:
- Problem: ChatGPT limited to pre-2021 data
- Solution: Retrieval Augmented Generation (RAG)
- Used LangChain technology
- Backend APIs: CoinGecko, DeFi Llama
- System selects appropriate API calls, processes JSON output, summarizes for humans
- Could answer live crypto questions (top coins by region, market trends)
- LangChain knowledge directly applied to subsequent John Deere internship
Career Philosophy and Advice
- Importance of experimentation and trying new things
- Value of diverse coursework beyond core major
- Personal projects and extracurricular involvement crucial
- Real-world application very different from classroom learning
- Success builds on success - projects led to more opportunities
- It's okay not to know your exact path initially
- Work experience can provide clarity for further education decisions
- Master's degree often required for applied sciences roles
Technical Skills and Learning
- Python, scikit-learn, feature engineering
- Natural Language Processing techniques
- TF-IDF analysis
- Vector databases and embeddings
- LangChain for automated LLM interactions
- Retrieval Augmented Generation (RAG)
- API integration and JSON processing
Future Plans
- Complete Master's degree while working full-time
- Focus on applied sciences rather than traditional software engineering
- Continue building on NLP and generative AI expertise
