Fault Lines & Safety Nets: How RAFFLES Finds the First Domino in Agent Failures
RAFFLES shows how structured, iterative evaluators can trace the first decisive fault in failed LLM-agent workflows, turning opaque failures into actionable diagnosis.
Cognaptus Insights
Industry thoughts, trends, and commentary.
From PDF to PI: Turning Papers into Productive Agents
Paper2Agent shows how research papers can become tested, MCP-backed execution interfaces—not just documents with better chatbots attached.
HyFedRAG: Caching Privacy into Federated RAG
HyFedRAG shows that federated RAG for sensitive healthcare data is less about one clever retriever and more about where retrieval, summarisation, privacy controls, and caching are placed.
Pareto on Autopilot: Evolving RL Policies for Messy Supply Chains
MORSE reframes supply-chain optimisation as a switchable portfolio of Pareto-efficient RL policies, with CVaR added for tail-risk-aware operations.
Graph and Circumstance: Maestro Conducts Reliable AI Agents
Maestro shows why reliable AI agents need graph-level redesign, not just better prompts, and how businesses should interpret the evidence.
Mind the Gap: How OSC Turns Agent Chatter into Compound Intelligence
OSC shows why multi-agent LLM systems need an orchestration layer that manages communication itself, not merely expert selection and final aggregation.
Model Portfolio: When LLMs Sit the CFA
A CFA benchmark shows why finance AI needs task routing, selective retrieval, and calculation checks—not one heroic model.
Parallel Minds, Shorter Time: ParaThinker’s Native Thought Width
ParaThinker argues that reasoning models need less single-track rumination and more native parallel exploration, with measurable gains on math benchmarks and a useful warning for enterprise AI design.
Plan, Then Rewrite: Why Explicit Intent Wins in Agent Workflows
RECAP shows why agent systems need an explicit intent layer between messy dialogue and downstream planning—not prettier summaries, but cleaner operational contracts.
Agreeable to a Fault: Why LLM ‘People’ Can’t Hold Their Ground
A mechanism-first look at why synthetic LLM personas can sound socially plausible while failing stricter tests of behavioural coherence.