LISCHKA.LI

Cinematography

Author: user

  • Fresh from AWS Innovate 2025 in Geneva and entrepreneurial side events in Geneva and Lausanne, I’m energized by insights on how businesses can leverage AI, serverless architectures, and a problem-first mindset to thrive in 2025. From technical deep dives to startup pitches, one clear message emerged: innovation isn’t just about adopting cutting-edge tech—it’s about solving real problems efficiently and adapting to change with purpose.

    1. Serverless Solutions: Reclaim Time for What Matters

    Imagine redirecting 75% of your resources to building your core business instead of maintaining infrastructure. Stephan Hadinger, AWS’s Director of Technology, shared a striking statistic: typical tech companies spend 70% of their efforts just “keeping the lights on.” Serverless solutions like AWS Lambda flip this equation. By charging only for the milliseconds your code runs and scaling automatically, Lambda frees businesses to focus on what matters most. Boris Flesch confirmed this in Lausanne. By avoiding the burden of building and managing servers, his team could concentrate on their core business, driving greater efficiency and innovation.

    Hadinger highlighted a real-world example: Bonque Edel slashed server costs from $10,000/month to just $2 by going serverless—a game-changer for startups, marketers, or designers who want to prioritize creating value over managing tech. This resonates with a broader point Hadinger made about legacy constraints. Just as Roman chariots’ 144cm width influenced railway tracks and even space shuttle boosters 2,000 years later, outdated IT choices—like clinging to Oracle databases—can limit progress today. Serverless offers a way to break free from such constraints.

    2. Build from Problems, Not Solutions

    A pitch event in Lausanne drove home a powerful lesson: don’t start with a product; start with a pain point. Cohen put it bluntly: “Solve a problem, and you have a business.” AWS Solutions Architect Anthony Baschenis reinforced this, noting that even small efficiencies—like cutting a 6-minute daily task to 3—can spark a viable business. This problem-first mindset applies across industries. In HR, AI tools reduce hiring bias and streamline candidate screening. For marketers, automating data analysis frees time for creative campaigns.

    Morgan Stanley’s 2025 AI trends report underscores this: AI reasoning models are enabling context-aware recommendations, optimizing everything from customer support to strategic planning. By focusing on real pain points, businesses can harness AI to deliver measurable value.

    3. Agentic Multi-Modal AI: The Next Frontier

    AWS Innovate showcased the transformative potential of agentic multi-modal AI, where systems combine diverse inputs—like text, images, or video—with the ability to act autonomously beyond their own boundaries. Jordi Porta and Jerome Bachelet emphasized that true power lies in models that integrate multi-modality with agentic capabilities. Amazon Bedrock, for instance, unifies access to over 170 models, enabling agents to adapt to changing conditions and execute complex workflows.

    Consider Amazon Q, a business expert powered by agentic plugins. In a demo, Q analyzed a CV against a job description, rated applicants, and suggested interview questions. With a Salesforce plugin, it created cases directly from queries. Meanwhile, Amazon Nova Act demonstrated agentic browsing, navigating Amazon’s website to add items to a cart, while Nova Sonic enabled voice-to-voice interactions with similar capabilities. These tools illustrate how multi-modal agents can query data, act on it, and deliver results where users already work—like Slack or e-commerce platforms.

    For businesses, this means solving problems with unprecedented flexibility. An orchestral agent on Bedrock, for example, can coordinate travel plans by balancing personal preferences (e.g., a vegetarian daughter) and business needs (e.g., flights and hotels), generating a tailored game plan. This is AI as a proactive partner, not just a tool.

    4. Disruption Requires Commitment to Change

    Nexthink’s mantra—“disrupt, reinvent, transform”—captures the essence of innovation. Pivoting isn’t about abandoning ideas; it’s about evolving with commitment and clear communication. Their advice? Build where users already are, like Slack integrations over complex UIs, and treat your business model as a living prototype. For entrepreneurs, this means staying cloud-native to avoid infrastructure burdens and focus on customer problems. As Nexthink emphasized, “AI is here to stay—use it for everything you do.”

    This aligns with insights from Pictet Asset Management’s Assad, who described using AI to streamline investment research and contract comparisons, and David Bloch’s focus on modern data strategies to accelerate business value. Disruption demands flexibility and a willingness to adapt.

    Looking Ahead: Business in 2026

    As AI becomes the “new electricity,” businesses in 2026 will need to fully embrace cloud-native architectures and agentic AI to stay competitive. Serverless solutions like AWS Lambda will further democratize innovation, enabling even small teams to scale without infrastructure burdens. Multi-modal agents, like those powered by Amazon Bedrock, will evolve to handle increasingly complex tasks—think orchestrating entire business workflows, from supply chain optimization to personalized customer experiences, with minimal human input. The problem-first mindset will remain critical: companies that identify and solve niche pain points, no matter how small, will carve out defensible markets. As Nexthink’s pivot journey suggests, flexibility and user-centric design will define success, with AI-driven tools embedded where users already work, like Slack or CRMs. By 2026, businesses that treat AI as a proactive partner—rather than a tool—will lead the charge in redefining industries.

    Four Action Points

    Geneva’s lessons boil down to asking better questions and acting decisively. Here’s how to move forward:

    1. Embrace serverless to cut costs and focus on growth.
    2. Solve real problems, no matter how small—AI reasoning makes it scalable.
    3. Leverage agentic multi-modal AI to act on data and deliver solutions where users are.
    4. Commit to change, treating AI as the new electricity to power every aspect of your business.

    As Dr. Andrew Ng said, “AI is the new electricity.” In 2025, it’s not just about adopting AI—it’s about using it to solve problems, act intelligently, and transform how we work.

    — Ramon Lischka

  • Untitled post 9097

    Purpose-Aligned Agile Video Marketing Workflow

    The “Purpose-Aligned Agile Video Marketing Workflow” is a structured, collaborative process that integrates a corporation’s mission-driven purpose with customer-centric insights, cross-functional teamwork, and rapid, data-driven video production. It emphasizes agility, technology, and simplified decision-making to create impactful video content efficiently, while managing resources and risks. The workflow ensures marketing teams and video producers deliver client-facing videos that resonate with audiences, achieve measurable goals, and adapt over time.

    The name draws from “The Ultimate Marketing Machine” (HBR, 2014) by Marc de Swaan Arons, Frank van den Driest, and Keith Weed—highlighting “Purpose-Driven Marketing” (mission alignment), “Agility and Speed” (rapid execution), and video as a key tool. “Workflow” reflects its actionable, step-by-step design for video production. source

    Why Use This Strategy?

    Alignment with Purpose

    Ensures videos reflect the company’s core mission, building authenticity and trust with audiences.

    Customer Focus

    Leverages data to target the right viewers with content they care about, boosting engagement.

    Efficiency

    Combines agility and streamlined decisions to deliver high-quality videos quickly, optimizing time and resources.

    Collaboration

    Integrates marketing, production, and other teams for cohesive output that supports broader goals.

    Adaptability

    Uses testing, feedback, and risk planning to keep content relevant and resilient in a fast-changing market.

    Scalability

    Relies on technology and lessons learned to streamline processes, making it repeatable and improvable over time.

    Who is it for?

    For companies that value purpose (to differentiate), need speed (to compete), and rely on video (to engage). It’s less suited for slow-moving industries or those without a clear mission or digital focus, but shines for dynamic, audience-driven organizations.

    Consumer Goods Companies

    Brands with a strong mission—like sustainability—can craft videos that reinforce their values and connect with customers emotionally.

    Tech Startups

    Fast-growing companies with limited resources can produce quick, smart videos to keep up with market changes.

    E-commerce Businesses

    Online sellers can create customer-focused videos to showcase products and drive sales effectively.

    Digital-First Consumer Brands

    Companies big on social media—like fashion or fitness—can make fast, platform-ready videos to stay relevant.

    Larger Companies with Multiple Teams

    Big organizations can align marketing, sales, and other teams for consistent, impactful video content.

    Creative Agencies

    Agencies can deliver tailored, efficient videos for diverse clients, balancing creativity and strategy.

    Step by Step Guide

    Step 1: Align on Purpose

    • Objective: Establish the video’s purpose tied to the corporation’s mission.
    • Marketing Team:
      • Share the mission and goals (e.g., “Promote sustainability with this video”).
      • Define the desired outcome (e.g., “Boost brand awareness”).
    • Video Producer:
      • Ask clarifying questions (e.g., “What’s the core message?”).
      • Draft a purpose statement (e.g., “Inspire sustainable choices”).
    • Together: Agree on the purpose as the guiding focus.
    • Time: 1-2 hours (kickoff meeting).
    • Tools: Client brief, mission statement, Zoom.

    Step 2: Understand the Audience

    • Objective: Identify the target audience and their needs.
    • Marketing Team:
      • Provide audience data (e.g., “Young, eco-conscious viewers”).
      • Note key behaviors (e.g., “They prefer short content”).
    • Video Producer:
      • Review data and ask for details (e.g., “What motivates them?”).
      • Create a viewer persona (e.g., “Eco-friendly young adult”).
    • Together: Confirm the persona and needs to target.
    • Time: 2-4 hours.
    • Tools: Analytics, social media insights, client research.

    Step 3: Gather Cross-Functional Input

    • Objective: Align the video with broader business goals.
    • Marketing Team:
      • Get input from other teams (e.g., “Sales wants a key feature shown”).
      • Pass it to the producer.
    • Video Producer:
      • Suggest how to include input (e.g., “I’ll highlight that feature”).
      • Raise concerns (e.g., “Will this shift the tone?”).
    • Together: Agree on essentials to include.
    • Time: 1-3 hours (meeting or emails).
    • Tools: Slack, project boards, team notes.

    Step 4: Brainstorm and Concept Quickly

    • Objective: Develop a rough video idea fast.
    • Marketing Team:
      • Give creative direction (e.g., “Make it upbeat”).
      • Set a quick deadline (e.g., “Concept by tomorrow”).
    • Video Producer:
      • Draft a storyboard or outline (e.g., “30-second customer story”).
      • Share for feedback.
    • Together: Refine the concept in one round.
    • Time: 4-6 hours.
    • Tools: Sketch tools, Google Docs, paper.

    Step 5: Plan with Data, Tech, and Resources

    • Objective: Shape the video with data and tech, considering budget.
    • Marketing Team:
      • Share past video data (e.g., “Short clips work best”).
      • Define platforms and budget limits (e.g., “$5,000 max”).
    • Video Producer:
      • Adjust based on data (e.g., “15-second cut”).
      • Propose cost-effective tech (e.g., “Stock footage instead of a shoot”).
    • Together: Lock in format, approach, and resource plan.
    • Time: 2-3 hours.
    • Tools: Analytics, editing software, budget sheet.

    Step 6: Build the Team

    • Objective: Assemble a lean, capable team within budget.
    • Marketing Team:
      • Outline needs (e.g., “Script and visuals”) and resources (e.g., “Use our editor”).
      • Approve team size and costs.
    • Video Producer:
      • Assign roles (e.g., “I direct, they shoot”).
      • Note gaps (e.g., “Need a freelancer?”).
    • Together: Finalize the team, keeping it cost-effective.
    • Time: 1-2 hours.
    • Tools: Email, contacts, budget tracker.

    Step 7: Produce a Test Video

    • Objective: Create a rough cut to test.
    • Marketing Team:
      • Pick the test audience (e.g., “Social followers”).
      • Set initial metrics (e.g., “Half should finish it”).
    • Video Producer:
      • Shoot and edit a quick version (e.g., “15-second teaser”).
      • Deliver for testing.
    • Together: Review and approve for testing.
    • Time: 1-2 days.
    • Tools: Camera, basic editing software, file sharing.

    Step 8: Test, Refine, and Mitigate Risks

    • Objective: Improve the video and plan for setbacks.
    • Marketing Team:
      • Test the video (e.g., “Post online, track views”).
      • Share feedback (e.g., “Client wants it faster”).
    • Video Producer:
      • Analyze results (e.g., “Cut the slow bit”).
      • Plan contingencies (e.g., “Backup shoot if rejected”).
    • Together: Agree on changes and a risk plan (e.g., “Extra edit time if needed”).
    • Time: 1-2 days (test + refine).
    • Tools: Social platforms, feedback forms, editing tools.

    Step 9: Finalize the Video

    • Objective: Produce the final video smoothly.
    • Marketing Team:
      • Approve the concept (e.g., “Good—just adjust pacing”).
      • Keep feedback quick (one round).
    • Video Producer:
      • Shoot and edit the final cut (e.g., “60 seconds, polished”).
      • Apply changes and finalize.
    • Together: Sign off on the final version.
    • Time: 2-5 days.
    • Tools: Pro equipment, editing software, branding guides.

    Step 10: Launch, Measure, and Iterate

    • Objective: Deliver, assess, and improve based on results.
    • Marketing Team:
      • Launch the video (e.g., “Post on social, website”).
      • Track KPIs (e.g., “10% engagement, 5K views”).
    • Video Producer:
      • Provide files in all formats (e.g., “MP4, vertical”).
      • Suggest tweaks (e.g., “More calls-to-action next time”).
    • Together: Review performance and plan iterations (e.g., “Version 2 with faster cuts”).
    • Time: 1-2 hours (launch), ongoing analysis.
    • Tools: Hosting platforms, analytics, email.

  • Understanding AI Model Quantization on Arch Linux

    AI models, particularly deep neural networks, often demand significant computational resources and memory, making them impractical for edge devices or lightweight systems. Quantization addresses this by reducing the precision of model weights and activations—e.g., from 32-bit floats to 8-bit integers—trading minimal accuracy for speed and efficiency. On Arch Linux, with its bleeding-edge tools, you can experiment with quantization techniques to optimize models. This guide introduces the core concepts and common quantization methods, tailored for an Arch environment.

    Prerequisites

    You’ll need a working Arch Linux system, basic Python knowledge, and familiarity with AI frameworks like PyTorch or TensorFlow. A pre-trained model (e.g., a PyTorch vision model) is helpful for testing. Access to a terminal and sufficient disk space for dependencies are assumed.

    Setting Up the Environment

    Install Python and PyTorch, a popular framework with built-in quantization support, along with pip for additional packages.

    sudo pacman -S python python-pip python-pytorch

    Verify PyTorch installation by checking its version in Python.

    python -c "import torch; print(torch.__version__)"

    For GPU support, install pytorch-cuda if you have an NVIDIA card and CUDA setup.

    sudo pacman -S python-pytorch-cuda

    Understanding Quantization Basics

    Quantization reduces the bit-width of numbers in a model. Full-precision models typically use 32-bit floating-point (FP32) for weights and activations. Quantized models might use 16-bit floats (FP16), 8-bit integers (INT8), or even lower, shrinking model size and speeding up inference. Three main approaches exist: post-training quantization (PTQ), quantization-aware training (QAT), and dynamic quantization.

    Post-Training Quantization (PTQ)

    PTQ applies quantization after training, converting a pre-trained FP32 model to a lower precision like INT8. It’s simple and doesn’t require retraining, but accuracy may drop slightly. Test it with a PyTorch script using a pre-trained ResNet18 model.

    import torch
from torch.quantization import quantize_dynamic
model = torch.hub.load('pytorch/vision', 'resnet18', pretrained=True)
model.eval()
quantized_model = quantize_dynamic(model, {torch.nn.Linear}, dtype=torch.qint8)
torch.save(quantized_model.state_dict(), 'resnet18_ptq.pth')

    This dynamically quantizes linear layers to INT8. Run it and compare model size.

    ls -lh resnet18.pth resnet18_ptq.pth

    Quantization-Aware Training (QAT)

    QAT simulates quantization during training, allowing the model to adapt to lower precision. It’s more complex but preserves accuracy better than PTQ. Here’s a minimal QAT example with a fake quantization step.

    import torch
import torch.nn as nn
from torch.quantization import prepare, convert
model = torch.hub.load('pytorch/vision', 'resnet18', pretrained=True)
model.train()
qconfig = torch.quantization.get_default_qat_qconfig('fbgemm')
model.qconfig = qconfig
prepare(model, inplace=True)
# Simulate training loop (not shown)
model.eval()
quantized_model = convert(model)
torch.save(quantized_model.state_dict(), 'resnet18_qat.pth')

    Insert a training loop with your dataset before converting. QAT typically yields smaller, faster models with less accuracy loss.

    Dynamic Quantization

    Dynamic quantization quantizes weights statically but computes activations dynamically at runtime. It’s lightweight and suits models with heavy linear operations. The PTQ example above uses this method—note the {torch.nn.Linear} specification.

    Comparing Quantization Effects

    Evaluate model size and inference speed post-quantization. Load both original and quantized models, then time a sample inference.

    import torch
import time
model = torch.hub.load('pytorch/vision', 'resnet18', pretrained=True)
model.eval()
quantized_model = torch.load('resnet18_ptq.pth')
input = torch.randn(1, 3, 224, 224)
start = time.time()
model(input)
print(f"FP32: {time.time() - start:.3f}s")
start = time.time()
quantized_model(input)
print(f"INT8: {time.time() - start:.3f}s")

    Smaller sizes (e.g., ~40MB to ~10MB) and faster inference (often 2-3x) are typical gains, though accuracy needs validation with your test set.

    Troubleshooting

    If quantization fails, ensure PyTorch supports your model’s layers (some custom ops may not quantize). Check for overflow errors with INT8—QAT can help. For GPU issues, verify CUDA compatibility or fallback to CPU.

    python -c "import torch; print(torch.cuda.is_available())"

    Quantization on Arch Linux empowers you to slim down AI models for deployment, balancing efficiency and precision with tools fresh from the repos.

  • Photography Backup System on Arch Linux via SSH

    Photography Backup System on Arch Linux via SSH

    Photographers rely on raw files and edited images as their lifeblood, making off-site backups essential. On Arch Linux, rsync over SSH paired with cron offers a secure, automated solution to mirror your photo library to a remote server. This guide configures a photography backup system optimized for large files, leveraging Arch’s lightweight tools and SSH’s robust security to protect your work.

    Prerequisites

    A remote server with SSH access and ample storage is required, along with basic terminal skills. Your photo collection—likely in ~/Photos or similar—should be ready to sync. Both local and remote systems need rsync; the remote server must also support SSH.

    Installing Core Tools

    Install rsync for file transfers, openssh for secure communication, and cronie for scheduling on your Arch system.

    sudo pacman -S rsync openssh cronie

    Activate cron to enable automated tasks.

    sudo systemctl enable crond
    sudo systemctl start crond

    Securing SSH Access

    For secure backups, set up SSH key authentication. Generate a key pair and transfer the public key to the remote server. Use a strong passphrase for added protection.

    ssh-keygen -t ed25519 -C "photo-backup"
    ssh-copy-id -i ~/.ssh/id_ed25519.pub user@remote-server

    Verify seamless access and lock down SSH by disabling password logins on the remote server’s /etc/ssh/sshd_config (set PasswordAuthentication no and restart sshd).

    ssh user@remote-server

    Prepare a remote directory, such as /backup/photos, with appropriate permissions.

    ssh user@remote-server 'mkdir -p /backup/photos && chmod 700 /backup/photos'

    Writing the Backup Script

    Create a script to sync your photo directory to the remote server, optimized for large raw files. This uses rsync’s compression and incremental transfers to save bandwidth.

    #!/bin/bash
# photo_backup.sh
SOURCE="$HOME/Photos/"
DEST="user@remote-server:/backup/photos/"
rsync -avzh --progress --delete -e "ssh -i $HOME/.ssh/id_ed25519" "$SOURCE" "$DEST"

    Save as ~/photo_backup.sh, make it executable, and test it. The -z flag compresses data, ideal for raw files, while –delete ensures the remote reflects the source.

    chmod +x ~/photo_backup.sh
    ~/photo_backup.sh

    Automating with Cron

    Schedule nightly backups at 1 AM by editing your crontab, logging results for monitoring.

    crontab -e

    Insert this line.

    0 1 * * * /home/user/photo_backup.sh > /home/user/photo_backup.log 2>&1

    Check cron’s status to ensure it’s operational.

    systemctl status crond

    Validating the Backup

    Confirm the backup’s integrity by comparing file counts or sizes.

    find ~/Photos/ -type f | wc -l
    ssh user@remote-server 'find /backup/photos/ -type f | wc -l'

    Simulate a sync to spot discrepancies without changes.

    rsync -avzh --dry-run --delete -e "ssh -i $HOME/.ssh/id_ed25519" ~/Photos/ user@remote-server:/backup/photos/

    Troubleshooting

    If transfers fail, review the log for rsync or SSH errors (e.g., key rejection). Test connectivity and permissions.

    cat ~/photo_backup.log
    ssh -v user@remote-server

    Monitor remote storage and adjust as your collection grows. This setup delivers a secure, efficient backup system for photographers on Arch Linux.

  • Claudia

    Apply to work with Claudia

    Claudia is an international model, actress, and advertising face.

    With over 10 years of experience in the fields of fashion, portrait, beauty, and acting roles, she has a broad range of expertise.

    She has worked with many talented photographers, and her career also includes appearances in Netflix productions and collaborations with classic modelling agencies.

    Claudia brings professionalism, routine, and reliability, coupled with a genuine passion for the modelling profession. She describes herself as authentic, natural, dependable, punctual, and straightforward – qualities that make her a reliable partner in any project.

    Claudia is looking forward to working with you!

  • Firmenevent Dübendorf