engblogs

Eight-week randomized controlled trial in Sierra Leone assesses Gemini's Guided Learning, an AI-powered pedagogical tool that augments teachers to raise math achievement, boost engagement, and promote Socratic, scaffolded learning.

A pragmatic look at how agentic coding and stronger type systems are redefining formal methods to scale verification, improve feedback, and push reliable software to new heights.

From initial skepticism to adoption, the piece argues that agentic coding can make formal methods as practical as type systems, easing verification bottlenecks and enabling stronger proofs in software development.

From pilots to production: how federal and regulated deployments build trusted AI agents through evidence-driven evaluation, task-specific rubrics, and clear shared accountability for risk.

Turning demos into reliable, production-grade AI for government and regulated industries, this post explains why trust, not just accuracy, matters, outlines an evaluation lifecycle (task, trace, outcome, rubric), and advocates a data-centric, risk-aware, human-in-the-loop approach.

Human-centered AI: how ethics, governance, and education intersect with computing to guide responsible AI deployment and alignment with societal values.

Examines the crucial role of human judgment, ethics, and education in guiding responsible AI and computing, focusing on AI alignment, governance, and the societal responsibilities of technology.

A technical exploration of building a self-sustaining, collaboratively shared object system by representing the entire live state as an Automerge document, testing merges across multiple users, and drawing lessons from mixed results.

Modular Cloud's five-stage, plugin-based routing pipeline (Prepare, Filter, Score, Pick, Execute) enables composable end-to-end LLM inference routing and showcases disaggregated prefill/decode and KV-cache-aware optimizations for low latency and scalability.

AI-powered biodefense blueprint for the intelligence age, detailing rapid threat detection, countermeasure development, and coordinated crisis response with safeguards and governance.

Significantly enhanced, dreaming-driven memory architecture for ChatGPT that continuously curates and updates user context to deliver fresher, more personalized conversations at scale.

Endava reimagines software delivery by embedding AI agents across engineering and business workflows, accelerating delivery and shaping an AI-native operating model.