Dev Blogs
ISE hackathon produced a reusable Japanese LLM evaluation pipeline with multi-turn, facet-based scoring, cross-provider model support, and publication-ready benchmarking outputs.
A multimodal LLM enrichment pipeline that extracts structured metadata from visual assets, constrains output to predefined values to minimize hallucinations, and uses a ground truth evaluation template to measure quality—all as a plug-and-play module.
How query decomposition separates metadata filters from visual intent to significantly improve media asset retrieval quality.
When distributing LLM calls across PySpark workers via mapInPandas, MLflow autolog silently fails. Here is how to fix it.
How we solved the challenge of passing conversational context between independently deployed agents using the A2A protocol's embedded context pattern, keeping domain agents stateless and secure.
When 100% of our prototype outputs were valid JSON but 0% met the data contract, we discovered the LLM was doing work that software should own. A field classification exercise and 4-pass pipeline brought schema compliance from 0% to 100%.
Real-world lessons from evolving a production chatbot into a coordinator-based multi-agent architecture, including performance trade-offs for enterprise-scale agent reuse.
A data-driven framework we use in enterprise deployments to decide between vector-only keyword and hybrid search, based on five measurable evaluation criteria.
A practical pattern for using GenAI research agents to produce reliable internal tooling by verifying assumptions with detection classes and a final checklist.
How to move AI-assisted development from ad-hoc experimentation to a coordinated team-wide capability using AGENTS.md and reusable skills with GitHub Copilot CLI
