Systems | Development | Analytics | API | Testing

This session takes a practical look inside LLM context windows and token consumption, exploring what happens when context enters a model - from tokenization, embeddings, attention, QKV, prefill, and decode to KV caching. It also examines how context windows are allocated and why simply increasing context length doesn’t always lead to better model performance.

This video shows how Astera ReportMiner extracts structured data from PDFs in two ways: a reusable template for consistent layouts, and an AI-driven pipeline for documents that vary. Working from a mix of PDFs, digital and scanned, that vary in layout and quality, you can build a custom extraction logic in ReportMiner: Whether a PDF arrives with a known layout or a new one, it runs through the same pipeline into the same output, with fewer manual exceptions and less setup.

Mocking for testing starts off easy, but once you scale to multiple teams and AI agents, handcrafted mocks become a serious form of technical liability. Instead of treating mocking as an individual software engineering task, shift your mindset to treat it as a platform engineering task focused on automation and continuously refreshed modern data. Watch to see how adopting technologies like traffic replay to simulate realistic backend sandboxes can transform your modern testing workflow!

Human biology is extraordinarily complex, and researchers often have only fragments of information to work with. Brian Martin compares it to looking at a skyscraper through a keyhole: you can see something clearly, but only a tiny piece of the whole. Recorded at EVOLVE26 Singapore, this episode of The AI Forecast brings Paul Muller together with Brian Martin, CTO of Applied AI at Cloudera and co-founder of Rare Hopes NFP, to explore what one of the world’s most data-intensive industries can teach us about AI and decision-making.

Discover how Onlyfy by XING, an all-in-one recruiting and applicant management solution partnered with ThoughtSpot, a leader in search and AI-driven analytics that empowers businesses with intuitive, self-service data experiences.

As companies adopt AI coding tools, code review processes are breaking down. Traditional compliance methods are slowing teams down, but human engineers aren't going to spend hours reading AI-generated low-level code forever. How will SOC 2 adapt to the era of AI-driven development? Drop your thoughts in the comments and subscribe for more tech insights! Learn more: speedscale.com.

This video shows how Astera ReportMiner extracts data from documents that don't share a common layout, with no template to design and no fields to mark. Traditional pattern matching depends on documents landing in the same shape every time. Invoices from hundreds of different suppliers don't, and this demo uses that as the test case: mixed layouts, multiple languages, multi-page files, and scans. The walkthrough covers.

Not everyone who needs a read on release readiness lives in Jira. This video demonstrates how the SmartBear MCP Server brings trusted Zephyr test data into Claude, so anyone who needs access to testing data gets it without switching context. A QE lead checking coverage, a release manager confirming a ship date, a developer following up on a fix, and any stakeholder who just wants a fast answer has what they need.

Manually cross-validating virtual card data across disconnected systems was a constant pain, until the procurement team built an AgentSpot agent to do it automatically. This video showcases a Virtual Card Intelligence agent that pulls PO and requisition data from Google Drive, cross-validates it against Spend Flow, flags expiring cards and utilization risks across 30/60/90 day bands, and sends personalized Slack outreach to card owners. Plus a full 9-tab Excel report covering KPIs, expiry, utilization, and reconciliation. The best part? This is all done in just 4-5 minutes.

The procurement team at ThoughtSpot built two AgentSpot agents to tackle manual, time-consuming workflows — and the results speak for themselves. Agent 1 – Coupa PR Approval Nudger: Automated purchase requisition reminders that once took hours of manual chasing in Coupa and Slack. Approval cycle time dropped from 6.62 days to 2.8 days. Agent 2 – Virtual Card Intelligence: Cross-validates PO and requisition data against Spend Flow to surface budget leakage, flag expiring cards, and send personalized Slack outreach to card owners. All without manual audits.