Downloaded 16 times
Uploaded byAll Things Open
Spacewalk 2026 - Presented at the IMAX in Raleigh, NC
Presented at Spacewalk 2026 Presentations include: AI Tooling in 2026 - Mark Hinkle, CEO/Founder @ Peripety Labs AI Governance - Nithya Ruff, Head of the Amazon Open Source Program Office/Chair Linux Foundation Board Severing the Tether: Why Cloud AI Can't Run the Real World - Robert Thelen, CEO & Co-Founder, LlamaFarm AI-powered voice agents and production-ready voice AI systems - Andre Lassiter, Managing Partner, MS3IT Business Solutions Tech Education in the Age of AI - Eli "the Computer Guy" Etherton, Silicon Dojo AI & Accessibility - Maria Lamardo, Sr. Accessibility Program Manager, GitHub AI & DevRel in 2026 - Director of Product Ecosystems (DoPE), Budhaditya Bhattacharya, Tyk
More Related Content
Similar to Spacewalk 2026 - Presented at the IMAX in Raleigh, NC
Spacewalk 2026 - Presented at the IMAX in Raleigh, NC
- 2. 1. AI Toolingin 2026 - Mark Hinkle, CEO/Founder @ Peripety Labs 2. AI Governance - Nithya Ruff, Head of the Amazon Open Source Program Office/Chair Linux Foundation Board 3. Severing the Tether: Why Cloud AI Can't Run the Real World - Robert Thelen, CEO & Co-Founder, LlamaFarm 4. AI-powered voice agents and production-ready voice AI systems - Andre Lassiter, Managing Partner, MS3IT Business Solutions 5. Tech Education in the Age of AI - Eli "the Computer Guy" Etherton, Silicon Dojo 6. AI & DevRel in 2026 - Director of Product Ecosystems (DoPE), Budhaditya Bhattacharya, Tyk 7. AI & Accessibility - Maria Lamardo, Sr. Accessibility Program Manager, GitHub Tonight’s Program: Seven Speakers, Out of this World AI
- 3. AI Tooling in2026 Mark Hinkle CEO, Peripety Labs Co-founder, All Things AI
- 4. AI Tooling in2026 What You Need to Know this year Mark Hinkle Co-Founder, All Things AI SpaceWalk 2026 | All Things Open
- 5. The Opportunity Create, Analyze,Execute Create Generate text, images, and video from simple descriptions instantly. Analyze Process complex datasets and find actionable insights in seconds. Execute Deploy autonomous agents to complete multi- step workflows for you. Get up to speed quickly with the free 14-day course at theaios.ai
- 6. The Landscape Real Risks,Real Solutions The Risks Data Privacy Your data could be exposed, shared, or misused by public models. Human Error AI amplifies mistakes and hallucinations at machine speed. Existential Threat The "Terminator" scenario: loss of control and alignment issues. The Solution "Education is our best defense." Verify every output before acting. Understand what AI can and cannot do. Adopt responsible usage policies.
- 7. Step 1 Master FrontierModels ChatGPT The Standard The go-to for general reasoning, creative writing, and daily tasks. Excellent all-rounder. Claude The Analyst Superior for coding, processing large documents, and nuanced writing. Microsoft Copilot The Enterprise Best for corporate environments with deep Office 365 and data integration. Gemini The Ecosystem Deeply integrated into Google Workspace (Docs, Gmail, Drive) for seamless workflows.
- 8. Step 2 Automate MenialWork 28% of your work week That's how much time the average professional spends on email. Most of it doesn't require your full attention. Tools That Actually Work Organization & Drafting Fyxer & SaneBox Smart filtering and organization. Superhuman Speed and AI-powered search. Autonomous Management Shortwave AI summaries and drafting. JACE Full autonomous email handling. "The Goal: Let AI triage and draft. You review and send."
- 9. Step 3: FromChat to Action AI Agents Execute Workflows Chatbots Talk. Agents Do. While chatbots answer questions, Agents autonomously perform multi-step tasks, browse the web, and use tools to complete work. Manus.im The general agent for research, coding, and complex automation. Open Source Stack LangChain for workflows, HuggingFace for models, Ollama for local deployment. Autonomous systems in action
- 10. Step 4: MultimediaCreation Create Stunning Assets Instantly Video Generation Sora 2, RunwayML Create cinematic video from text prompts. AI Editing Veed.io Edit, caption, and repurpose video automatically. Image Generation Google Nano Banana, Midjourney Generate high-fidelity visuals for any context.
- 11. Next Steps Join Usat All Things AI 2026 March 23–24, 2026 Durham, NC Carolina Theatre & Convention Center Register at allthingsopen.org Free Course at theaios.ai "The best time to learn AI was yesterday. The second best time is today."
- 12. AI Governance: Your Roleas a User Nithya A. Ruff Director, Amazon Open Source Program Office
- 13. Brand Presentation Template AmazonConfidential AI is not an overnight success
- 14. Brand Presentation Template AmazonConfidential In the last few years, conditions became ripe for AI to become viable
- 15. Brand Presentation Template AmazonConfidential AI has generated so much excitement
- 16. Brand Presentation Template AmazonConfidential And everyone is talking about it
- 17. Brand Presentation Template AmazonConfidential And there is also fear uncertainty and doubt
- 18. Brand Presentation Template AmazonConfidential As with any technology, we worry about what can go wrong
- 19. Brand Presentation Template AmazonConfidential The good news is….
- 20. Brand Presentation Template AmazonConfidential There is a lot of good governance work being done
- 21. Brand Presentation Template AmazonConfidential Regulations, Product Testing, Policies, Guardrails
- 22. Brand Presentation Template AmazonConfidential 7 important things Users can do to benefit from AI and reduce risks
- 23. Brand Presentation Template AmazonConfidential One: Educate yourself
- 24. Brand Presentation Template AmazonConfidential Two: Use and experiment
- 25. Brand Presentation Template AmazonConfidential Three: Do not share personal or company data
- 26. Brand Presentation Template AmazonConfidential Four: Trust but verify the outputs
- 27. Brand Presentation Template AmazonConfidential Five: Know your organization’s AI Policy
- 28. Brand Presentation Template AmazonConfidential Six: Use AI ethically and responsibly
- 29. Brand Presentation Template AmazonConfidential Seven: Follow secure practices Watch out for Scams
- 30.
- 31. To know moreabout how we work with AI check out https://aws.amazon.com/ai/
- 32. We all havea role to play in using AI responsibly
- 33. Brand Presentation Template AmazonConfidential Thank you 3 3
- 34. Severing the Tether AI atthe Edge Matt Hamann CTO, LlamaFarm.dev
- 48. Voice AI inthe Real World Andre Lassiter MS3IT Business Solutions
- 49. SPACEWALK 2026 |ALL THINGS OPEN Voice AI in the Real World Andre Lassiter | MS3IT Business Solutions
- 50. The $1,500 Problem Saturdayafternoon. Peak rush. Every technician busy. A property management company needed 20 tablets repaired before Monday. The phone rang. And went to voicemail. They called the next shop. 20 tablets x $75 avg repair = $1,500 gone in four rings 62% CALLS GO UNANSWERED DURING PEAK HOURS 85% OF THOSE CALLERS NEVER CALL BACK $126K AVERAGE ANNUAL REVENUE LOST PER SMB
- 51. The Tooling CaughtUp TWO YEARS AGO 2-3s RESPONSE LATENCY <80% SPEECH RECOGNITION ACCURACY TODAY 100-300ms VOICE-TO-VOICE LATENCY 95%+ SPEECH RECOGNITION ACCURACY
- 52. The 300ms Rule Humanconversation operates in a 300-500ms response window. Go beyond 800ms, you lose the conversation. Beyond 1.5s, it feels like a phone menu. Deepgram STT ~150ms Groq LLM ~200ms ElevenLabs TTS ~75ms Total Pipeline OPTIMAL
- 53. Two Sides ofVoice AI INBOUND Customer calls you. Reactive architecture. > Qualify leads > Route to departments > Capture demand 24/7 > Book appointments OUTBOUND AI calls the customer. Proactive architecture. > Appointment confirmations > Reactivation campaigns > Follow-up sequences > Survey collection Different prompting psychology. Different compliance requirements. Same underlying stack.
- 54. Architecture by Scale SMALL <10Employees Solo plumber, 3-person HVAC shop Full replacement for after-hours and overflow. Owner is on job sites all day. AI answers 24/7. $78K Annual value from 5 extra jobs/week MEDIUM 10-100 Employees Regional med spa, multi-truck roofing Hybrid deployment. AI handles overflow + outbound grunt work. Humans focus on complex conversations. 500+ Reactivation calls per month ENTERPRISE 100+ Employees Regional hospital, national insurance First-line triage. Handles the 60% of calls that don't need a human. 14% Increase in issue resolution
- 55. The Stack 1 STT-> Audio to text. Whisper, Deepgram, Vosk. 2 LLM -> Intent + response generation. GPT-4, Claude, Llama. 3 TTS -> Text to audio. Chatterbox, Kokoro, ElevenLabs. 4 Telephony -> SIP trunks, Twilio, WebRTC. 5 Orchestration -> Turn detection, interruption handling, state.
- 56. Open Source Frameworks LiveKitAgents Powers ChatGPT Voice Mode. WebRTC- based. Python & Node.js SDKs. 100K+ devs 3B calls/yr $45M Series B Pipecat Daily.co's framework. Vendor-neutral with 40+ AI integrations. Built-in voicemail detection. 5K+ GitHub stars 600+ forks Python 3.10+ Vocode Get a prototype in 20 lines of Python. Great for proving concepts fast. Quick start MIT License
- 57. STT & TTSOptions // SPEECH-TO-TEXT Whisper large-v3 1.55B params | 1M hrs training | 99+ languages | ~7% WER Whisper Turbo 6x faster | 809M params | 1-2% accuracy trade-off Deepgram Nova-3 ~150ms latency | 54% WER reduction | Commercial Vosk Offline-capable | Edge deployments // TEXT-TO-SPEECH Coqui XTTS-v2 Voice cloning from 6s sample | 17 languages | Non-commercial Chatterbox MIT license | 350M params | 23 languages | Emotion control Kokoro 82M params | Apache 2.0 | <$1 per million chars
- 58. The Hard Parts Integrationis 70% of the work Voice AI is easy. Getting it to do something useful after it talks - that's where projects die. ServiceTitan, Housecall Pro, Boulevard, Zenoti, Salesforce, HubSpot - each has its own auth flow, rate limits, data models. Real example: Booking an HVAC appointment = 5 API calls minimum -> technician availability + service area + job type + duration + customer history Latency compounds across the pipeline STT (100-500ms) + LLM (350ms-1s+) + TTS (75-200ms) = 800ms-1.5s in real conditions. Add distributed infra: STT in us-east-1, LLM in us-west- 2, TTS in eu-west-1. Each hop = 30-70ms. Fix: Co-locate services in the same region. Same AZ if possible. Every ms matters.
- 59. The Hard Parts(continued) Scale changes everything 50 calls/day -> 500 -> 5,000. Each level breaks the previous architecture. Connection pooling, webhook queues, horizontal scaling, circuit breakers, graceful degradation. Backpressure, thundering herd, cold start - all the boring distributed systems problems apply. OpenAI Realtime API = $20/hr Great for demos. Incredible latency. But at contact center scale: 10K calls/day x 5 min avg = 833 hrs = $16,660/day = $500K/month Cascaded pipelines (STT->LLM->TTS) are dramatically cheaper at scale. People problems > Tech problems 70% of AI value depends on process and adoption, not technology. If your team doesn't trust the AI, they route around it. If your processes are broken, AI breaks them faster. The technology is ready. The question is whether your organization is.
- 60. The Money isFlowing $2.4B 2024 $47.5B 2034 34.8% Compound Annual Growth Rate 7x VC investment increase in 2 years ($315M in 2022 -> $2.1B in 2024)
- 61. The Stack isReady ORCHESTRATION LiveKit / Pipecat SPEECH-TO-TEXT Whisper / Deepgram TEXT-TO-SPEECH Chatterbox / Kokoro LANGUAGE MODEL Your choice Andre Lassiter MS3IT Business Solutions "Every missed call is $200-500 walking out the door." Want to talk architecture, tooling, or deployment patterns? Find me after this.
- 62. Let's Connect Questions? Wantto discuss Voice AI for your business? https://andre.ms3it.com Scan the QR code on screen or visit the URL directly
- 63. The Stack isDead: Long Live the Stack Eli "the Computer Guy" Etherton Silicon Dojo
- 64. The Stack isDead Long Live the Stack Tech Education in the Age of AI
- 65. I’m Eli theComputer Guy
- 66.
- 67. AI is Justa Stack
- 68. Are Tech JobsDead?
- 69.
- 70. Where is YourInfrastructure
- 71. Self Harm onCorporate AI
- 72. Risk of OpenAIBusiness Model
- 73. What Do We“Do”?
- 74.
- 75. Not Understanding Reasonfor Job or Tasks
- 76. Not Understanding Valueof Company
- 77.
- 78. Teach Business andSolving Problems Through the Lens of Technological Solutions Why Does This Work Matter? What is On the Line for the Success/ Failure of Work What are the Systems Considerations for Building Solutions What “AI” Functionality Can Be Used to Increase Efficiency
- 79. “AI” Isn’t Goingto Take Your Job… Not Keeping Up With the Steady March of Progress Will…
- 80. Building the new DevRelPlaybook Budhaditya Bhattacharya, Tyk Director of Product Ecosystems (DoPE),
- 81. Building the new DevRelPlaybook Spacewalk 2026 AI-augmented DevRel in 2026
- 82. “AI is notgoing to replace DevRel!”
- 83. But is itgoing to replace you and me…
- 84. Budhaditya Bhattacharya Director ofProduct Ecosystem, Tyk Certified: Real human India, Singapore, USA (Durham) ● Product education, community engagement, and open-source ecosystem expansion ● Chairperson of the OpenAPI Initiative (BGB) ● Horror movies & musicals
- 85. “Every DevRel storycombines curiosity, connection & authenticity — things that make us deeply human.”
- 86. Reactions to change Curiosity& excitement Apathy Fear & skepticism
- 87. Reactions to change Curiosity& excitement Fear & skepticism
- 88. What makes agood DevRel Adaptability
- 89. Adapting to change Understandingthe changes Updating our DevRel playbook
- 90.
- 91. ● Vibe coders ●Developers ● Product managers ● Designers ● Finance teams ● Marketing teams 1. Our audience is changing Builders ● App developers ● Software developers ● Engineers ● Solutions Architects Developers
- 92. 2. Content consumptionis changing AI agents + Humans ● Docs ● Blogs ● Videos ● Demos Humans
- 93. 3. The metricsare changing AEO ● Domain authority ● Rankings ● Impressions ● Traffic ● Conversions SEO ● Mentions ● Citations ● Placements ● Referral traffic ● Conversions
- 94.
- 95. ● Start withwhy, who and what ● We provide the strategy, AI amplifies it ● Become a content curator ● Community Insights & engagement ● Identify trends & report on the business impact ● Human in the loop The AI-augmented DevRel Playbook
- 96. Reframing AI +DevRel What will it help me unlock What is it going to take away from me
- 97.
- 98. AI & Accessibility MariaLamardo Sr. Accessibility Program Manager, GitHub
- 99. AI and Accessibility MariaLamardo Senior Accessibility Program Manager
- 100. DORA Survey Insight 90% ofdevelopers who responded to the 2025 DORA survey are using AI for code generation.
Editor's Notes
- #13 Many scientists have been working for decades on AI. And in the last decade many facts came together to make it happen.
- #21 The AI act Security testing of products for data, model hallucination etc./ human in the loop, verifying and being actable for decisions made by tool Guard rails in tools – to prevent you from making mistakes.
- #25 Avoid entering sensitive personal, financial, health, or confidential work information into public AI tools unless explicitly approved and contractually protected. Assume prompts and files may be stored or used to improve models unless the provider clearly says otherwise. Use strong, unique passwords and enable multi-factor authentication on AI services and related accounts to prevent account takeover and data exposure.
- #26 You are the expert. Review the outputs, is it right, Check sources, don’t just use it without verifying or making it your own Code, writing, apps etc. Treat AI responses as drafts or suggestions and always verify important facts, legal points, medical advice, or financial guidance with reliable primary sources or professionals. Watch for confident but wrong answers, outdated information, and missing context. Be cautious about copying AI-generated text, code, or images directly into production systems, contracts, or public posts without review for accuracy, bias, plagiarism, and policy compliance.
- #27 And respect the policy. What tools you can use, what data you can share etc. Someone in your org has done their homework and has determined what you can and cannot use. Respect that. At work or school, follow your organization’s AI policy and stick to approved tools, which are more likely to have vetted privacy, security, and retention practices. When in doubt, ask whether a tool is approved before using real customer or internal data. Review providers’ terms, data policies, and opt-out options so you understand how your prompts, files, and outputs are stored, shared, or used to train models.
- #28 As with any tool – the internet, the mobile phone etc. The capabilities can be used for good or for bad. And it is your responsibility as a user to make sure you are using it ethically and responsibly. We e Avoid using AI for harmful purposes such as harassment, discrimination, automated deception, or spreading deepfakes and misinformation. Many institutional guidelines explicitly classify these as unacceptable uses that can violate law and policy. Consider people who could be affected by your AI-assisted content (for example, students, coworkers, or audiences online) and disclose AI use when it materially shapes what they see or rely on.
- #29 Be skeptical of emails, messages, or content that may be AI-generated for phishing or social engineering; verify identities before sharing information or sending money. Check URLs, sender details, and request legitimacy, especially if there is urgency or pressure. Keep devices, browsers, and AI apps updated, and prefer services from reputable providers that document their security practices and allow reporting of abuse.
- #30 Is any discussion today complete without a discussion of AI. It is the single biggest disruption we are seeing in all of our lives today. From the developer to the business owner to the consumer at home, it is changing how we search, research, create code or talk to our doctors or plan a vacation. While mush of the underlying software used in creating models and apps is open source – think pytorch, Jupiter, langchain, python AI poses a number of challenges to open source in this third decade. First – we all need to come together on what open source AI means and how we comply with this definition. I commend the OSI and the LF AI and Data on their work in this area. We have more work to do on adopting and complying with the definition. Second AI is generating a lot of code which can overwhelm maintainers and create challenges in copyrights fair use etc. Projects are taking stances on how they accept AI generated code, and remember that you as a developer should still be in charge. To guide, review and approve what a coding assistant suggests. Understanding what is in our AI consumption is another areas we are working to understand - so what we did with software needs to be expanded to capture all the diff components that will come in when we consume a model or a dataset or an AI app. Lastly, there are not enough open source models in the world. And those are needed to engage a broader community of developers in fine tuning, recreating and making it work for their country or their use case. We are limited to a few proprietary models with less transparency and a chance to influence the bias and ethics on. Shaping the future of open source AI: Discuss the community's role in advocating for transparency, ethical data use, and responsible practices as AI continues to evolve. Example: Mention the move toward more visible AI governance and tools.
- #32 As users, we must actively invest in the projects we rely on. As companies, we must move beyond passive consumption and build cultures of contribution. And as a community, we must champion the principles of sustainability, security, and ethical innovation for the decades to come."
- #33 What was important is – having a policy and strategy that fits the company’s goals and culture Building open source processes and thinking into the processes and mechanisms of the company – so it is easy and integrated to do it right. Building an OSPO to create bridges with the external community – to engage, learn and give back
- #35 Good evening! My name is Matt Hamann and I’m a co-founder and the CTO of LlamaFarm. Tonight, let’s talk about severing the tether. Cutting our dependence on Cloud AI and moving our AI workloads to the edge.
- #36 You’re building an agent… … But the problem is that every API call is a dependency you don’t control, a bill you have to pay, and a prayer that everything is just going to work. Sometimes this is fine!
- #37 Over the holidays, a family member left some sourdough starter in my fridge.
- #38 Take stoplights for example…
- #39 Hospital Emergency Dept First responders
- #41 So, why is now the right time for Edge AI? Well, I think we’re at an inflection point… Models shrank… Hardware caught up… Tooling is improving… Getting started…
- #42 …is easy! Choose a model, select a quantization, pick a runtime, grab a framework, maybe a database, get everything connected in a way that makes sense, and deploy to production. And hope it all works! What about errors? Hallucinations? How do you monitor this thing? You basically have to become an AI expert. It’s overwhelming. It’s complex. It’s almost impossible to figure out how all these pieces all fit together.
- #44 You may recall that DeepSeek popularized the term “mixture of experts” where different parts of a model’s “brain” activate depending on the current context.
- #46 To get started, create a llamafarm project file and start the dev server. Llamafarm takes care of the rest: pulling models, optimizing for your hardware, and updating as your project evolves, all while providing an OpenAI compatible API wherever you need it.
- #47 So join us in severing the tether. GitHub.com/llamafarm Support us by starring the repo. Try it out, give us feedback. Be part of our community and contribute back. Together, we can build the future of edge AI.
