Technical Articles & Systems Research
Engineering teardowns, mathematical formulations, and production architectural playbooks designed for both novice intuition and intermediate/senior engineering rigor.
Why OCR Breaks Your RAG (And How ColPali Fixes It)
Why traditional text parsers scramble complex multi-column balance sheets and tables. A practical deep dive into VLM Markdown generation vs. native patch-level Late Interaction with ColPali.
Evaluating RAG Beyond Vibe Checks: A Production Testing Playbook
Moving beyond manual eyeballing. How to isolate retrieval failures from generation hallucinations using the Ragas Triad, calibrated LLM judges, and continuous regression gates.
The 1-Keyword Trap: Mastering Hybrid Search & Reciprocal Rank Fusion
Why pure vector embeddings fail on single-keyword deltas, how Reciprocal Rank Fusion ($k=60$) mathematically reconciles candidate lists, and how cross-encoders eliminate semantic collisions. Includes interactive playground!
When RAG Fails: Building Self-Correcting Pipelines with LangGraph
Architecting stateful retrieval loops. Features an autonomous document grader, query rewrite agent, and fallback search degradation trees using LangGraph.
Why Most Multi-Agent Swarms Fail in Production
Connecting 5 agents in a peer-to-peer swarm explodes message complexity at $O(N^2)$. How to architect hierarchical supervisors, decouple tools via the Model Context Protocol (MCP), and manage immutable state checkpoints.
Debugging Multi-Agent Systems: The 3 AM Runaway Loop
Post-mortem of a $1,240 runaway agent deadlock. How to diagnose cascading hallucinations, mathematically bound loop convergence via Lyapunov functions, and instrument OpenTelemetry distributed trace spans.
Securing Autonomous Agents: Stopping the Confused Deputy Attack
The Trojan Resume exploit in production. How indirect prompt injections weaponize privileged tool execution agents, and how to enforce Bell-LaPadula information flow control and 3-tier capability sandboxing.
Human-in-the-Loop for AI Agents: Cryptographic Dual-Key Approvals
When autonomy must halt for irreversible actions. How to build durable state suspension, asynchronous approval queues, and cryptographic HMAC-SHA256 signature validation with LangGraph interrupts.