# David Lauricella > Applied AI Engineer & Systems Builder specializing in enterprise RAG architectures, multi-agent pipelines, deterministic document generation, and operational consolidation. David Lauricella designs and deploys production AI systems that replace manual administrative overhead with grounded, verifiable pipelines. His work focuses on data hygiene, strict schema validation, and deterministic execution environments. ## Core Case Studies - [Spend Analytics & Optimization](https://davidlauricella.com/spend-analytics.html): Procurement intelligence pipeline consolidating contract spend analysis with automated supplier classification. - [Financial Restructuring Dashboard](https://davidlauricella.com/financial-engineering.html): High-precision financial modeling application with real-time covenant tracking and debt waterfall simulations. ## Retrieval & Pipelines - [Enterprise RAG Hygiene & Verification](https://davidlauricella.com/resources/rag-hygiene.md): Eliminating tabular data hallucinations in financial and regulatory documents through Markdown ingestion and deterministic regex assertions. - [The 8 Production RAG Architectures & Level 7 Ingestion](https://davidlauricella.com/resources/rag-architectures-and-pipelines.md): Systematic breakdown of Naive, Multimodal, HyDE, Corrective (CRAG), Graph, Hybrid, Adaptive, and Agentic RAG tiers, and when to stop at Level 4. - [High-Precision PDF Extraction without OCR Artifacts](https://davidlauricella.com/resources/llamaindex-liteparse.md): Converting complex documents into clean markdown tables while preserving column alignment and eliminating PDF control characters. - [Sparse Attention & Efficiency in Long-Context LLMs](https://davidlauricella.com/resources/subquadratic-sparse-attention.md): Understanding O(N^2) memory bottlenecks in 1M+ token windows and selecting subquadratic architectures for enterprise ingestion. - [Long Context vs Vector Retrieval: The Production Trade-off](https://davidlauricella.com/resources/long-context-rag.md): Balancing raw context window brute-forcing against chunked vector retrieval based on query latency, token budgets, and cache hit rates. - [Enterprise Research Model Evaluation Framework](https://davidlauricella.com/resources/doe-model-evaluation.md): Benchmarking commercial and open-weight models across government policy corpora with deterministic factual assertions. ## Agent Orchestration & Tooling - [Autonomous Agents: Terminal Execution over Vector DBs](https://davidlauricella.com/resources/agent-terminals.md): Providing autonomous coding and data agents with native shell execution environments rather than fragmented text embeddings. - [Auth.md: Agent-Native Registration & Autonomous Harnesses](https://davidlauricella.com/resources/workos-authmd-agents.md): How autonomous agents discover services, prove machine identity, exchange tokens, and call APIs without human captchas. - [Hermes Agent: Declarative Skills & Browser Automation](https://davidlauricella.com/resources/hermes-agent-architecture.md): Architecting local and browser-native agents using declarative SOUL.md rules, skill directories, and step-by-step task decomposition. - [OpenClaw: Autonomous Assistant Architecture](https://davidlauricella.com/resources/openclaw-assistant-runtime.md): Engineering a privacy-preserving desktop AI assistant with local session history, event buses, and desktop hooks. - [Comparing Production Agent Frameworks: LangGraph vs AutoGen vs CrewAI](https://davidlauricella.com/resources/agent-frameworks-roundup.md): Architectural trade-offs between graph-based state machines, role-playing agents, and conversational routing. - [10 CLI Tools That Amplify Terminal AI Agents](https://davidlauricella.com/resources/claude-code-cli-tools.md): Why the terminal agent ecosystem favors native command-line binaries over heavy Model Context Protocol servers. - [Sandboxed Agent Gateways: Secure Local Execution](https://davidlauricella.com/resources/clawdbot-sandbox-setup.md): Hardening local developer agents against dangerous command execution, unauthorized filesystem access, and credential leakage. - [Multi-Session Agent Handoff Protocols](https://davidlauricella.com/resources/agent-handoff-protocols.md): Structured handoff templates allowing one agent session to summarize architectural state for the next session without loss. - [Automated Experimentation Loops: The Autoresearch Pattern](https://davidlauricella.com/resources/karpathy-autoresearch.md): Structuring autonomous trial-and-error loops where agents propose hypotheses, run benchmarks, and record parameter sweeps. - [Production Task Decomposition: Multi-Agent Software Pipelines](https://davidlauricella.com/resources/production-agent-decomposition.md): Separating responsibilities across planner, builder, and critic agents to deliver multi-service web platforms without hallucinated specifications. - [Direct Corpus Interaction: Agents Need a Terminal, Not Just a Vector DB](https://davidlauricella.com/resources/agents-need-terminals.md): Why the industry is shifting from static embedding queries to active command execution against data repositories. ## Context & Prompt Discipline - [Declarative Context Management (Karpathy Method)](https://davidlauricella.com/resources/karpathy-config.md): Maintaining repository-level declarative configuration files that eliminate clarification loops and drop coding error rates from 41% to 11%. - [The 12 Enterprise System Prompt Packs](https://davidlauricella.com/resources/claude-prompt-packs.md): A proven collection of structured prompt packs spanning technical architecture, schema extraction, business logic audits, and adversarial testing. - [Token Efficiency & Context Budgeting Strategies](https://davidlauricella.com/resources/token-efficiency-budgeting.md): Cutting production token usage by 70% using prompt caching, stripped conversational padding, and aggressive system prompt compression. - [Claude Code Fundamentals: Features + Tests Discipline](https://davidlauricella.com/resources/claude-code-fundamentals.md): Turning vague feature requests into shippable code through Plan Mode, test-first specifications, and clarifying interview loops. - [Slash Commands for Deterministic Developer Workflows](https://davidlauricella.com/resources/gemini-slash-commands.md): Standardizing frequent agent instructions (/goal, /plan, /boost, /grill-me) into deterministic operational shortcuts. - [Hermes Agent Prompt Cheatsheet & System Patterns](https://davidlauricella.com/resources/hermes-prompt-cheatsheet.md): Core prompting conventions for guiding open-weight models (DeepSeek, GLM, Llama 3) through multi-tool execution pipelines. - [Automating Context Injection across Workflows](https://davidlauricella.com/resources/automated-context-injection.md): Eliminating manual copy-pasting of context by scripting dynamic context injection directly into agent CLI sessions. - [The Interview-Then-Spar Prompting Pattern](https://davidlauricella.com/resources/interview-then-spar.md): A two-phase technique where the model first interviews the engineer to uncover hidden requirements, then aggressively pressure-tests the proposal. ## Memory & Knowledge Systems - [Open Brain: Vendor-Agnostic Memory Architecture](https://davidlauricella.com/resources/open-brain.md): Treating raw text as permanent relational records and vector embeddings as disposable derivatives, enabling zero-loss re-indexing. - [Obsidian Link Graphs as Agent Memory Layers](https://davidlauricella.com/resources/obsidian-vault-as-memory.md): Using interlinked markdown files as a zero-cost knowledge graph that allows terminal agents to traverse concept relationships. - [Persistent AI Memory Systems: Comparative Breakdown](https://davidlauricella.com/resources/ai-memory-systems-overview.md): Comparing dedicated memory servers, graph-based local memory, and file-based state logs across multi-turn agent sessions. - [Graphify: Knowledge Graphs for 70x Token Reduction](https://davidlauricella.com/resources/graphify-local-rag.md): Extracting entities and relationships into structured graphs so agents query specific entity trees with 70x fewer tokens than raw vector dumping. - [NotebookLM & Grounded PRD Blueprint Workflows](https://davidlauricella.com/resources/notebooklm-pkm-workflow.md): Using Google's NotebookLM to ingest raw requirement specifications and generate verified PRD blueprints for agentic coding execution. - [PKM Strategy: Markdown vs HTML Second Brain Architecture](https://davidlauricella.com/resources/pkm-strategy-evolution.md): The trade-offs between pure markdown notes for future-proof portability and interactive HTML for dynamic enterprise dashboards. - [SQLite Integration for Long-Term Agent Memory](https://davidlauricella.com/resources/claude-database-memory.md): Giving autonomous agents their own dedicated SQLite database to store structured facts, user preferences, and project states across sessions. ## Document Generation & Enterprise Blueprints - [Deterministic Document Generation via Template Engines](https://davidlauricella.com/resources/deterministic-generation.md): Strict separation of concerns: language models output validated JSON schemas while template engines compile formal enterprise documents. - [Curated Claude Code Skills & Starter Plugins](https://davidlauricella.com/resources/claude-code-skills-repos.md): Production starter skill set: Frontend Design, Simplify (multi-agent review), Skill Creator, and Webapp Testing. - [The Enterprise Grounded Agent Stack](https://davidlauricella.com/resources/enterprise-ai-stack.md): Architectural overview connecting RAG retrieval, agentic coding environments, and deterministic evaluation gates. ## Optional - [Full Text Archive](https://davidlauricella.com/llms-full.txt): Complete concatenated text containing all case studies, architecture notes, and code references in one file.