From fa2644d9baa1e0542ecc52ce04b51c8f0b97f105 Mon Sep 17 00:00:00 2001 From: Anton Akusok Date: Sun, 9 Mar 2025 18:34:55 +0200 Subject: [PATCH] Script draft --- 10.Video/script.md | 131 +++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 131 insertions(+) create mode 100644 10.Video/script.md diff --git a/10.Video/script.md b/10.Video/script.md new file mode 100644 index 0000000..6d298a2 --- /dev/null +++ b/10.Video/script.md @@ -0,0 +1,131 @@ + +# Video script + + +## AI Script + +Creating a 5-minute video that presents a new software architecture using Amazon's "working backwards" principle involves focusing on customer needs and clearly explaining how the architecture addresses these needs. Here’s a structured guideline for your video: + +We are internal architects, trying to sell the idea to our own business. + +**How to make:** ask AI to create small chunks of a video. + + +## [Opening Scene – 0:00-0:30] + +Visuals: Engaging animation or graphics showing a customer-focused environment + +Narrator: +"Welcome to [Your Company/Team Name], where innovation is driven by your needs. Today, we unveil our latest software architecture designed to transform how you [specific benefit related to the architecture]." + +- intro to AI +- people working --> machines working, people service machines + + +## [Scene 1: Define Customer Needs – 0:30-1:00] + +Visuals: Customers facing challenges, infographics highlighting pain points + +Narrator: +"Before we started designing, we asked ourselves: What are the biggest challenges our customers face today? We listened to [specific problems] and identified key areas for improvement: [list 2-3 major pain points]." + +Quote from CEO: "Change is hard on its own, and keeping the grading quality consistent while hanging is even harder." + +On-screen text: Key customer pain points + +- static scale up 5-10x, impossible with current resources +- dynamics of scale up fast enough (hiring, reduced bandwidth while growing, smooth transition) +- keep consistent quality while changing the business +- paying for the scale up, margins are thin at the moment +- heard about AI but how to make it happen consistently, reliably, maintaining the control? + + +## [Scene 2: Present the Solution – 1:00-2:30] + +Visuals: Transition to technical diagrams, flowcharts of the new architecture, animations demonstrating core features + +Narrator: +"Our response is a groundbreaking software architecture that [briefly describe the core function]. By leveraging [mention any technologies, e.g., microservices, cloud computing], we've built a system that is [mention key attributes like scalable, secure, efficient]." + +Technical Explanation: + +Scalability: "Our architecture scales effortlessly with your business, thanks to [specific technology or method]." +Security: "Security is foundational, with [describe security features]." +Performance: "Optimized for speed and reliability, our system delivers [describe performance benefits]." +Interview Clip: +Lead Architect (on-screen): +"This architecture allows us to integrate seamlessly with existing systems and adapt to future needs, ensuring your operations are always at the forefront of innovation." + +-------- + +- general approach 1: AI is good at simple tasks, bad at complex tasks +- general approach 2: divide and conquer, split grading into simple tasks and handle each one separately +- general approach 3: ensure highest grading quality locally (per-task) and globally (per-applicant) + +- tasks: a question or a grading aspect of task 2 +- keep a larger pool of questions or grading dimensions than necessary, to enable smooth rotation +- AI task workflow: + 1. introduce a new task or a new grading dimension + 2. deploy to a small share of students and grade manually + 3. deployment eval: run AI system in parallel until it reaches "X" performance metric; do prompt tuning at same time + 4. switch grading to AI; if AI system never stabilizes then goto 6 + 5. continuous eval: keep manually grading a small part of submissions, check AI system performance over time + 6. sunset: obsolete question or poor AI performance --> replace question with another +- global system eval: + 1. send 5% students to manual evaluation + 2. 4% are with the largest AI uncertainty (sample from the decision boundary of Certifiable Inc "function") + 2. compare per-task performance and whole system performance + 3. investigate differences, add new questions of grading aspects that open up these differences +- question variations: propose few different formulations for the same question, grade with the same AI system +- fraud detection + + +## [Scene 3: Customer Impact and Testimonials – 2:30-3:30] + +Visuals: Happy customers using the new system, graphs showing improvement metrics + +Narrator: +"The impact? [Specific benefit] for our customers. Here’s what a few of them have to say." + +Customer Testimonial Clips: + +Customer 1 (on-screen): "With [software name], we've seen a [specific improvement], making our workflow smoother and more efficient." + +Specific improvements: + +- prompts to check for specific parts in the solution document + + +Customer 2 (on-screen): "The transition was seamless, and the support has been exceptional." + +- transition with shifting one question at a time +- use historical evals to test model performance + + +## [Scene 4: Architecture Details & Trade-Offs – 3:30-4:30] + +Visuals: FAQ list, roadmap timeline with future features + +Narrator: +"We know you have questions. Here are some we’ve anticipated and addressed:" + +On-screen text: + +"How does it integrate with existing platforms?" +"What is the cost structure?" +"What support is available?" +Narrator: +"And we’re not stopping here. Our roadmap includes [mention upcoming features or enhancements]." + + +## [Closing Scene – 4:30-5:00] + +Visuals: Company logo, contact information, call to action + +Narrator: +"Thank you for joining us on this journey. We’re excited to continue working with you to create the future of [industry or specific area]. For more information, visit [website/contact info]. Let’s build the future together." + +This structure ensures that the video remains customer-focused, aligning with the "working backwards" principle while providing enough technical depth to convey the significance and capabilities of the new software architecture. + + +- info about our team