NVIDIA AI Podcast – Episode 283
"How Anyone Can Build Meaningful AI Without Code"
Guest: Shanaya Levin, Co-founder & CEO, Impromptu AI
Host: Noah Kravitz
Release Date: December 17, 2025
Episode Overview
This episode explores the democratization of artificial intelligence development. Shanaya Levin, co-founder and CEO of Impromptu AI, discusses her journey from big tech to pioneering no-code (and “mixed-code”) AI solutions. The conversation centers on how Impromptu empowers both technical and non-technical users, especially in enterprise environments, to build, deploy, and trust high-quality AI applications—without relying on deep AI expertise or heavy-duty machine learning teams. The discussion also touches on accessible design, the evolution of AI tooling, optimizing for accuracy, the importance of trust and "provable AI," and the broader impact on technology, business, and career pathways.
Shanaya Levin's Background and the Genesis of Impromptu AI
- Shanaya's Tech Path:
- Studied business and computer science
- Worked on developer tools at Google, including Google Home and Android [01:02]
- Roles at eBay, Cloudflare, Docker, and as head of product at a Series C startup
- Launched and sold CodeSee—an early generative coding assistant utilizing GPT-2 beta [02:03]
- Traveled the world post-acquisition, then met Dr. Sean Robinson and co-founded Impromptu
- Memorable quote:
"I've been doing this a really long time, helping people to understand their code... and then Coty got acquired and I decided to take some time off and travel around the world." (Shanaya, [02:03])
- Personal highlight: working and inventing alongside Dr. Sean Robinson, computational physicist and co-founder [03:43]
Initial Vision & Journey to Impromptu AI
- Origin Story:
- Impromptu didn’t start as a plan to launch a large tech company.
- Began by helping friends/networks manually build better, more accurate AI applications.
- Pivotal realization:
"The infrastructure to build AI applications across all of our customers is exactly the same. I wonder if we can agentically write this AI application." (Shanaya, [06:10])
- Dr. Sean initially thought “AI that builds AI” wasn’t possible... then, 20 minutes later: “Well, maybe.” [06:15]
- Fast forward: now have an AI specialized in building actual AI applications for clients.
Who Does Impromptu Serve?
- Works primarily with businesses—mid-size and large enterprises, but also innovative smaller teams. [06:56]
- Typical client: one with an existing SaaS platform, seeking to “transform into an AI-native company” (vs. starting from scratch).
-
Their technology enables the integration of advanced AI without needing to discard legacy codebases.
"We've invented five new pieces of tech, from adaptive context engines to infinite memory to our optimization core..." (Shanaya, [07:37])
Making AI Accessible (to All)
- Broadening Access:
- Mission: "Make AI accurate and accessible." [09:09]
- Not just for "less technical" users—recognizes even experienced devs are often new to generative AI.
"People have been developing for 20 years and this is a new technology that we're all learning at the same time." (Shanaya, [09:09])
- Approach:
- UI asks guided questions; supports users step-by-step through building apps.
-
Offers co-build model—Impomptu’s experts work alongside clients' teams for education and support. [10:07]
> "Every user has someone to guide them so it doesn’t feel like we just leave you in the lurch." (Shanaya, paraphrased [10:19])
- Ensuring Production Readiness:
- “I can get you a chatbot in probably 10 minutes, it has all the production ready pieces that you need.” [14:08]
- True challenge at scale is managing complexity and accuracy with massive data sets, interconnected workflows, and governance requirements.
Industry Trends & Types of Applications
- Physical AI:
- "Huge trend in physical AI—how do you move things in physical space?" (Shanaya, [12:09])
- Specialized & Custom Data Models:
- The frontier is customized applications, memory/context management, and provable/trusted AI.
- Recent launch: custom data models directly inside their builder for highly accurate, domain-specific apps. [12:34]
- Examples:
- Mom-and-daughter team building financial literacy tools
- CPG brand using AI to optimize ocean plastic recycling into activewear [16:46]
"How do I spin up a new app, a new city across the country to do the exact same thing? … That’s very custom data you’re not going to get from a general model. You need their system and models..." ([17:08])
Technical Deep Dive: Performance, CUDA, and Optimization
- Mixed-Code Approach:
- Impromptu bridges the gap between no-code and pro-code; describes itself as “mixed-code.” [20:33]
- Leverage NVIDIA CUDA:
- Uses CUDA libraries for embedding/classification, offloading heavy math to GPUs, boosting both performance and efficiency.
"All the heavy math natively runs on GPUs instead of slowly on CPUs. That really gives us two big wins: performance… and efficiency..." (Shanaya, [21:12])
- Uses CUDA libraries for embedding/classification, offloading heavy math to GPUs, boosting both performance and efficiency.
- Bridging the AI Research–Enterprise Gap:
- “Today you don’t need a big machine learning team to build something sophisticated. CUDA helps us with that hard math.” ([22:09])
- Self-serve for non-technical pros, with option to have Impromptu handle complex builds and integrations.
The 98% Accuracy "Secret Sauce" & Provable AI
- Accuracy Is Task-Based:
- Moves beyond model benchmarks to focus on actual enterprise task success rates.
“Our accuracy is actually redefining that success, which is task success. ... We allow the user to define what success means to them...” ([23:42])
- Moves beyond model benchmarks to focus on actual enterprise task success rates.
- Optimization Engine:
- Real-time optimization across prompts, models, data, and evals toward user-defined goals.
- Two modes:
- Manual: For technical users to tinker and optimize
- Automatic: For less-technical users, the system automatically iterates to achieve high task accuracy (often 98%+)
- Dashboards & Trust:
- Transparent, live metrics show accuracy/progress toward user-defined criteria.
- Commitment to “provable AI”—users see what the system is doing and can audit, review, or roll back actions.
“If you can’t see it, feel it, touch it... We subscribe to something called provable AI.” ([26:17])
Barriers, Diversity, and the Future of Tech Skills
- Personal Experience as a Woman in Tech:
- Shanaya shares her early career anxieties and her determination to ensure fear doesn't prevent anyone from pursuing tech ambitions.
-
“I don’t want women or kids ... to feel like you have an idea and you can’t execute it.” ([29:03])
- Emphasizes that now, more than ever, technical barriers are coming down:
“That barrier is fully down, right? Everybody is running.” ([30:38])
- Changing Pathways & Critical Skills:
- Coding skills remain valuable, but core strengths are critical thinking, systems thinking, and adaptability.
“Computer science is not about coding. It’s about cognition, thinking in systems, critical thinking... breaking problems down into small shippable chunks.” ([32:22])
- Future CS education should emphasize understanding, orchestration, and interpretation of AI-generated systems—not just syntax or manual coding.
- Coding skills remain valuable, but core strengths are critical thinking, systems thinking, and adaptability.
Forward-Looking Insights & Opportunities
- Biggest Potential for Impact:
- AI has the power to “fully remake our entire world”—not just tweak existing apps.
- Key: as accuracy, dexterity, and data capabilities improve, every industry will undergo change as big as the shift from web to mobile.
"The real opportunities are in fully remaking our entire world... every industry will shift exactly the same way we did for mobile..." (Shanaya, [35:54], [37:12])
- Barrier to Change:
- Many current roadblocks are human/political—not technical.
“The problems are human problems that we keep creating, but ... we could really make sure to make the world a truly better place.” ([37:52])
- Many current roadblocks are human/political—not technical.
Notable Quotes & Timestamps
- "I've been doing this a really long time, helping people to understand their code." – Shanaya [02:03]
- "You want an AI that builds AI?" – Dr. Sean (quoted by Shanaya) [06:15]
- "People have been developing for 20 years and this is a new technology we're all learning at the same time." – Shanaya [09:09]
- "Today you don't need a big machine learning team to build something sophisticated." – Shanaya [22:09]
- "Our accuracy is actually redefining that success, which is task success." – Shanaya [23:42]
- "If you can’t see it, feel it, touch it, in a theoretical sense … proving to the user that we're doing what we say we're doing is of the utmost importance." – Shanaya [26:17]
- "Computer science is not about coding. It's about cognition… critical thinking skills." – Shanaya [32:22]
- "The real opportunities are in fully remaking our entire world." – Shanaya [35:54]
- "The problems are human problems that we keep creating... but if we moved out of the way, we could really make sure to make the world a truly better place." – Shanaya [37:52]
Key Takeaways
- AI development is getting dramatically more accessible—not just for technical users but for domain experts and entrepreneurs across the board.
- The real meaning of “no-code” is broader than drag-and-drop: it’s about systems that guide, optimize, and support the user’s intent.
- Trust and transparency (“provable AI”) are as essential as raw capability for enterprise adoption.
- Future tech careers (and education) should focus as much on systems thinking, orchestration, and creativity as on traditional computer science or programming skills.
- AI-enabled transformation parallels the shift to mobile in its radical potential to reshape business and society.
Resources & Where to Connect
- Website: Impromptu AI
- LinkedIn: Shanaya Levin
- TikTok: @shaneya (for behind-the-scenes on building an AI company)
- Podcast Home: NVIDIA AI Podcast
