Podcast Summary: Embracing Digital Transformation
Episode #279: Embracing the Power of Small Language Models
Host: Dr. Darren Pulsipher
Guest: Chris Carter, CEO of Approyo
Date: July 21, 2025
Episode Overview
This episode explores the practical and strategic rise of Small Language Models (SLMs) as an alternative to massive Large Language Models (LLMs) in public sector digital transformation. Host Darren Pulsipher and guest Chris Carter dig into how SLMs offer more focused, efficient, and scalable solutions, especially for enterprise and SAP environments, by leveraging precise data, improved security, and reduced hallucination risks. The conversation moves from the evolution of AI to current hands-on best practices, with a strong emphasis on people, process, and technology.
Key Discussion Points & Insights
1. Origin Stories & the Evolution of AI
- Early Journey in Tech:
- Chris Carter recounts his beginnings with computers in the 1980s, from the Commodore Vic 20 to enterprise mainframes.
“I started on a Commodore Vic 20 in 1986... learning as much as I could. By the 1990s, I was already playing with what we now call AI.” (Chris Carter, 01:22)
- Both host and guest reminisce about programming on cassette tapes and the passion for tinkering. (03:22–03:49)
- Chris Carter recounts his beginnings with computers in the 1980s, from the Commodore Vic 20 to enterprise mainframes.
- AI Over the Decades:
- AI existed since the 1960s, but hardware limitations kept it niche. The leap came only when compute power caught up.
“Even when the hardware could keep up, the lift to actually program in AI was pretty substantial, absolutely massive.” (Darren Pulsipher, 04:08)
- AI existed since the 1960s, but hardware limitations kept it niche. The leap came only when compute power caught up.
2. Why Small Language Models?
- From Too Big to Just Right:
- The industry’s focus shifted from massive LLMs (with billions or trillions of parameters) to smaller, more practical models that solve targeted problems efficiently.
“I don’t need to boil the ocean… I just need to boil that database. I don’t want it to contaminate, I don’t want it to hallucinate… If I can do that with a smaller subset, so I start getting deeper with a smaller subset and SLM.” (Chris Carter, 05:14)
- The industry’s focus shifted from massive LLMs (with billions or trillions of parameters) to smaller, more practical models that solve targeted problems efficiently.
- Efficiency & Precision:
- SLMs can be run on standard laptops (x86 environments), bypassing the need for high-end GPUs, making the tech more accessible and reducing costs (06:34–07:26).
“You are not dumbing down. You are just simply consolidating on what you need to focus on.” (Chris Carter, 06:44)
- SLMs can be run on standard laptops (x86 environments), bypassing the need for high-end GPUs, making the tech more accessible and reducing costs (06:34–07:26).
3. Use Cases for SLMs: Focus on SAP
- Technical vs. Functional Applications:
- Technical: Code generation and analysis are safer and more relevant with SLMs focused only on the organization’s actual codebase.
“No SAP system is exactly the same... If I start leveraging others’ code, I’m really going to do some damage to [my] landscape.” (Chris Carter, 09:17)
- Functional: SLMs can empower HR and line-of-business users to query data (e.g., for vacation policies) in plain language, receiving real-time recommendations without custom code.
“All they have to do is ask it the question… Can I take time off?” (Chris Carter, 13:12)
- Technical: Code generation and analysis are safer and more relevant with SLMs focused only on the organization’s actual codebase.
4. Data Hygiene: Cleaning Up for SLMs
- The Need for Continuous Data Cleansing:
- Cleansed and fine-tuned data is vital for effective SLM deployment, as organizations shouldn’t rely on outdated or dirty datasets.
“If you’ve got a six terabyte database, something’s probably wrong. Start cleaning your data. Fine tune that data... That’s where the little jewels and nuggets [are].” (Chris Carter, 14:38)
- Cleansed and fine-tuned data is vital for effective SLM deployment, as organizations shouldn’t rely on outdated or dirty datasets.
- Ongoing Maintenance:
- Data entropy is inevitable—persistent cleaning is non-negotiable.
“You have to constantly be cleaning your data and it’s not a one-time and done… There’s this law called the law of entropy. Everything moves to a state of chaos.” (Darren Pulsipher, 16:07)
- Data entropy is inevitable—persistent cleaning is non-negotiable.
5. Building Adoption: Evangelizing SLMs in Organizations
- Education Through Demonstration:
- Show-and-tell sessions, lunch-and-learns, and prototyping encourage peer learning and spur experimentation.
“I’ll bring my laptop, I’ll bring some components… Some people just aren’t going to get into it… But then you start bringing over [others] to our side, and all of a sudden you see that light bulb kick on.” (Chris Carter, 19:22)
- Show-and-tell sessions, lunch-and-learns, and prototyping encourage peer learning and spur experimentation.
- Empowering Tinkerers:
- Encourage team members to explore use cases, choose SLMs based on needs (e.g., text, code, Q&A), and fine-tune for specific scenarios.
“I let my team members… start having the ideas, start getting into it… I want them to be inquisitive.” (Chris Carter, 21:37)
- Encourage team members to explore use cases, choose SLMs based on needs (e.g., text, code, Q&A), and fine-tune for specific scenarios.
6. Moving from Experimentation to Institutionalization
- Reusable Precision:
- SLMs should be built once, reused across departments or clients, and tailored for specific, high-value tasks.
“The future of AI isn’t necessarily about being bigger… It’s about building with precision.” (Chris Carter, 26:10)
- SLMs should be built once, reused across departments or clients, and tailored for specific, high-value tasks.
- Still Need Engineers—But Differently:
- The role shifts to system thinkers, not just coders.
“I’m not necessarily hiring junior developers anymore. I’m hiring that next level up, those individuals that can think more about what I need to build, how I need to build, rather than just keyboarding something.” (Chris Carter, 28:01)
- The role shifts to system thinkers, not just coders.
7. Educating & Sourcing the Next Generation
- Evolving Academic Curricula:
- Schools must adapt to teach not just coding, but system thinking and tinkering with AI as a toolset.
“I need thinkers who are tinkerers… I want them to have everything at their nose. I can’t have them just knowing the Encyclopedia Britannica. I need them to know the globe. I need every piece of information. But then I also need those individuals… to review what [AI] has come up with and bring out the best of it.” (Chris Carter, 29:37)
- Schools must adapt to teach not just coding, but system thinking and tinkering with AI as a toolset.
Notable Quotes & Memorable Moments
-
On the shift from LLMs to SLMs:
“I don’t need a boiling ocean… I just need to boil that database.”
— Chris Carter, 05:14 -
On the necessity of data hygiene:
“Garbage in is garbage out… It’s cleanliness in is cleanliness out.”
— Chris Carter, 17:36 -
On engaging teams:
“I’m a show and tell guy.”
— Chris Carter, 19:22 -
On the future of engineering and AI:
“The future of AI isn’t necessarily about being bigger… it’s about being smarter.”
— Chris Carter, 26:10 -
On skills for tomorrow’s workforce:
“I need thinkers who are tinkerers. Just like I need the Walt Disney folks, I need those people who are going to think about it and then start tinkering with it.”
— Chris Carter, 29:37
Timestamps for Key Segments
- Chris Carter’s Origin Story: 01:22–03:49
- The Evolution of AI & Democratization via LLMs: 04:00–05:14
- Why SLMs? Focus vs. Scale: 05:14–07:26
- Constraining & Using SLMs Practically: 07:26–09:17
- SAP Use Cases (Technical vs. Functional): 09:17–13:14
- Data Hygiene & Ongoing Cleaning: 14:38–16:44
- Adoption Strategies & Team Enablement: 19:22–20:32
- Getting Started with Models & Deployment Advice: 21:27–23:54
- From Experimentation to Production: 26:10–27:46
- Skills and Education for the AI Era: 28:01–29:37
Closing Thoughts
This episode provides a practical, clear-eyed look at the emergence of Small Language Models as a game-changer in enterprise digital transformation. Carter and Pulsipher’s hands-on insights and relatable anecdotes make the case for precision, efficiency, and continual learning in both technology and people. The conversation is consistently optimistic about AI’s role—but stresses that smarter, not just bigger, is the way forward.
