# Fern Rivero: public portfolio # Fernando (Fern) Rivero I’m Fernando Rivero. I turn enterprise AI strategy into working systems. With 12+ years in enterprise technology, I design the architecture, build the agents, and connect the results to the business. My work spans autonomous software delivery, enterprise knowledge, ServiceNow, and AI cost optimization. Based in Miami, Florida. Source: https://www.iamfern.com/about ## Focus - Generative AI strategy - Multi-agent systems - ServiceNow platform modernization - RAG and semantic context optimization - AI cost modeling - Build-vs-buy evaluation - Field CTO advisory ## Connect - https://www.linkedin.com/in/getfern - https://www.instagram.com/unferngettable_/ - https://www.threads.com/@unferngettable_ Contact: https://www.iamfern.com/#contact --- # Engineering projects - [Autonomous Software Delivery](https://www.iamfern.com/content/autonomous-sdlc.md): Built an agentic delivery system that turns production issues into competing fixes, tested pull requests, and reviewed releases. - [Enterprise AI Assistant](https://www.iamfern.com/content/enterprise-ai-assistant.md): Built a specialist AI assistant across 40,000+ ServiceNow documentation pages and cut per-request cost from about $2.19 to $0.04–$0.06. - [Enterprise Knowledge Platform](https://www.iamfern.com/content/enterprise-knowledge-platform.md): Built an enterprise knowledge workspace with six ingestion pipelines and 226,142+ indexed chunks across four sources, powered by Qdrant. - [ServiceNow Service Modeling](https://www.iamfern.com/content/csdm-inference.md): Built a multi-agent system that turns ServiceNow incident history into a CSDM service model and a 13-section report. - [Security Review Agents](https://www.iamfern.com/content/security-review-agents.md): Combined seven scanning tools with eight specialist AI reviewers to investigate security findings and filter noise before issues reach engineers. - [Meeting-to-Backlog Planner](https://www.iamfern.com/content/meeting-to-backlog.md): Built a planning pipeline that turns meeting decisions into prioritized delivery stories, refining 350 proposals to 215 across ten transcripts. - [AI Product Scorecard](https://www.iamfern.com/content/ai-leadership-dashboard.md): Built a leadership scorecard that connects AI adoption, model spend, infrastructure cost, and modeled value so leaders can decide where to invest. - [MAS Agents Research 1.0](https://www.iamfern.com/content/mas-agents-v1.md): Cyberpunk-themed dual-mode research orchestration system using specialized AI models for deep-dive analysis and map-reduce synthesis. - [MAS Agents Code](https://www.iamfern.com/content/mas-agents-code.md): Autonomous coding platform featuring a 7-agent ensemble with true parallel execution and consensus-based conflict resolution. - [ServiceNow Expert Agent](https://www.iamfern.com/content/servicenow-expert.md): Production-grade multi-agent system managing 12 ServiceNow domains with circuit breakers, checkpointing, and tenant isolation. - [Montana Video](https://www.iamfern.com/content/montana-video.md): Government relations platform processing legislative video with multi-modal analysis, diarization, and vector search. - [Guestlists Miami](https://www.iamfern.com/content/guestlists-miami.md): VIP nightlife management system with a public-facing AI concierge, enforcing strict write-only security models. - [Reddit Email Digest](https://www.iamfern.com/content/reddit-digest.md): LangGraph pipeline synthesizing daily and weekly Reddit newsletters using dual-model abstraction and structured parsing. - [Auto Job Applier](https://www.iamfern.com/content/auto-job-applier.md): Resilient Selenium-based bot with pluggable AI adapters (DeepSeek, OpenAI, Gemini) for automated job applications. - [Project DREA](https://www.iamfern.com/content/project-drea.md): Privacy-first emotional intelligence platform using Vertex AI RAG to analyze relationship dynamics and detect manipulation patterns. - [Pocket Officer](https://www.iamfern.com/content/pocket-officer.md): Real-time Florida statute violation identifier using Vertex AI Search grounding and structured output validation. - [Semantic Deduplication](https://www.iamfern.com/content/semantic-dedup.md): CPU-optimized semantic clustering engine achieving ~34% token reduction for RAG pipelines via SimHash and graph clustering. - [YouTube Deep Dive](https://www.iamfern.com/content/youtube-learning.md): Precision analysis tool for technical presentations, extracting deep architectural patterns from video transcripts. - [Raggity Chat POC](https://www.iamfern.com/content/raggity-chat.md): Workspace utilizing OpenAI Assistants API with file_search and Google Drive integration for collaborative RAG. - [CourtListener MCP](https://www.iamfern.com/content/court-listener-mcp.md): Specialized MCP server for querying legal opinions and filings via the CourtListener API. --- # Services - [Agentic AI & Delivery](https://www.iamfern.com/content/genai-strategy.md): Build agents that investigate issues, use tools, and prepare tested changes for review. - [AI Strategy & Cost](https://www.iamfern.com/content/ai-tool-evaluation.md): Compare the options. Measure quality and cost. Build the business case before the commitment. - [Enterprise Knowledge & ServiceNow](https://www.iamfern.com/content/enterprise-ai-innovation.md): Connect company knowledge to useful answers, with scoped access and traceable sources. --- # Autonomous Software Delivery Built an agentic delivery system that turns production issues into competing fixes, tested pull requests, and reviewed releases. Source: https://www.iamfern.com/projects/autonomous-sdlc Role: Architect & Engineer Contribution: Original agent orchestration and delivery tooling, integrated with self-hosted open-source infrastructure. ## architecture - Telemetry intake and issue deduplication feed a self-hosted Gitea agent hub. - Two isolated worktrees generate competing implementations; an independent judge scores correctness, scope, style, and tests. - A bounded repair pass addresses test failures. Pull requests retain the review and promotion gate. ## Results & scale - One automatic repair turn per run - Two candidate implementations scored across four review dimensions ## stack - LangGraph - Azure AI Foundry - Gitea - GlitchTip - Python --- # Enterprise AI Assistant Built a specialist AI assistant across 40,000+ ServiceNow documentation pages and cut per-request cost from about $2.19 to $0.04–$0.06. Source: https://www.iamfern.com/projects/enterprise-ai-assistant Role: Architect & Engineer Contribution: Designed and built the specialist orchestration, retrieval, and infrastructure on Open WebUI. ## architecture - Parallel specialists use LangGraph and a Claude Agent SDK orchestration path. - Tool allowlists, concurrency limits, and distributed session locks bound each request. - Source retrieval and citation editing support grounded responses. ## Results & scale - 40,000+ documentation pages indexed - Per-request cost reduced from about $2.19 to $0.04–$0.06 with the Agent SDK orchestration path ## stack - LangGraph - Claude Agent SDK - Azure AI Foundry - Redis - MCP --- # Enterprise Knowledge Platform Built an enterprise knowledge workspace with six ingestion pipelines and 226,142+ indexed chunks across four sources, powered by Qdrant. Source: https://www.iamfern.com/projects/enterprise-knowledge-platform Role: Architect & Engineer Contribution: Designed and built the ingestion pipelines, retrieval integrations, client extensions, and deployment on Open WebUI. ## architecture - Separate pipelines ingest documentation, training, delivery assets, architecture references, problem records, and support cases. - Deterministic chunk IDs make ingestion resumable without duplicate vectors. - Retrieval checks access grants and returns named sources to dedicated personas. ## Results & scale - 226,142+ chunks across four counted sources - Six ingestion pipelines and three grounded personas ## stack - Open WebUI - Qdrant - Azure AI Foundry - Python - PostgreSQL --- # ServiceNow Service Modeling Built a multi-agent system that turns ServiceNow incident history into a CSDM service model and a 13-section report. Source: https://www.iamfern.com/projects/csdm-inference Role: Architect & Engineer Contribution: Original worker, reducer, and report-generation architecture. ## architecture - Four workers analyze business, technology, foundation, and infrastructure domains. - A deterministic reducer merges results without another model call. - Checkpoints resume interrupted runs; five synthesizers produce separate report sections. ## Results & scale - 500-record processing chunks - Four domain workers and a 13-section report ## stack - Python - Azure AI Foundry - ServiceNow - CSDM --- # Security Review Agents Combined seven scanning tools with eight specialist AI reviewers to investigate security findings and filter noise before issues reach engineers. Source: https://www.iamfern.com/projects/security-review-agents Role: Architect & Engineer Contribution: Original specialist orchestration, finding review, and issue integration around existing scanning tools. ## architecture - Seven static analysis and dependency tools run before model analysis. - Eight CWE-scoped agents review distinct security domains. - A judge considers reachability, existing controls, and test context before findings are filed. ## Results & scale - Seven deterministic scanning tools - Eight specialist agents with an 80% default confidence gate ## stack - LangGraph - Python - Semgrep - Bandit - GitHub --- # Meeting-to-Backlog Planner Built a planning pipeline that turns meeting decisions into prioritized delivery stories, refining 350 proposals to 215 across ten transcripts. Source: https://www.iamfern.com/projects/meeting-to-backlog Role: Architect & Engineer Contribution: Original extraction, chronological reconciliation, and backlog-generation pipeline. ## architecture - Transcripts are extracted concurrently, then reconciled in meeting order. - Later decisions update earlier proposals instead of creating contradictions. - Story writing runs in parallel; RICE scoring scopes the first delivery phase. ## Results & scale - 350 proposals reduced to 215 on the same ten-transcript input - Approximately 13 minutes of end-to-end agent runtime ## stack - LangGraph - Python - Azure AI Foundry - ServiceNow --- # AI Product Scorecard Built a leadership scorecard that connects AI adoption, model spend, infrastructure cost, and modeled value so leaders can decide where to invest. Source: https://www.iamfern.com/projects/ai-leadership-dashboard Role: Architect & Engineer Contribution: Original analytics ingestion, cost attribution, and dashboard implementation. ## architecture - Normalizes product analytics and Azure Cost Management data. - Each metric carries a status such as measured, estimated, or insufficient data. - Cost attribution separates model usage from retrieval and hosting. ## stack - Python - React - Azure Cost Management - Recharts --- # MAS Agents Research 1.0 Cyberpunk-themed dual-mode research orchestration system using specialized AI models for deep-dive analysis and map-reduce synthesis. Source: https://www.iamfern.com/projects/mas-agents-v1 Role: Architect ## architecture - Dual-mode: Agentic Deep Dive + Map-Reduce Military Swarm - 6 specialized models (Planner, Researcher, Reasoner, Drafter, Validator) - Self-healing Engineer Agent generates Python tools at runtime ## sophistication - Runtime AST validation & security scanning - Semantic tool memory via ChromaDB - Vector-ID dispatch pattern (57% cost savings) ## scale - Processes 50-100+ web sources per query - 73% speed improvement via parallel dispatch ## stack - LangGraph - ChromaDB - Python - DeepSeek - Llama 3.3 --- # MAS Agents Code Autonomous coding platform featuring a 7-agent ensemble with true parallel execution and consensus-based conflict resolution. Source: https://www.iamfern.com/projects/mas-agents-code Role: Lead Engineer ## architecture - 7-agent ensemble (Architect, Coder, Tester, etc.) - Parallel execution via LangGraph Send API - Consensus voting protocol (Clarify/Concern/Vote) ## sophistication - Bounded ReAct loops (max 25 iterations) - Dual model tiers (Standard vs Premium routing) ## stack - LangGraph - Docker - Python - Claude 3.5 - Gemini --- # ServiceNow Expert Agent Production-grade multi-agent system managing 12 ServiceNow domains with circuit breakers, checkpointing, and tenant isolation. Source: https://www.iamfern.com/projects/servicenow-expert Role: Principal Engineer ## architecture - 18 specialized LangGraph agents - PostgreSQL checkpointing for session resumption - Per-user ChromaDB memory isolation ## scale - 354,977 LOC base - 10 concurrent asyncpg connections - Supports multi-tenancy via env override propagation ## stack - ServiceNow - LangGraph - PostgreSQL - MCP - Python --- # Montana Video Government relations platform processing legislative video with multi-modal analysis, diarization, and vector search. Source: https://www.iamfern.com/projects/montana-video Role: Lead Developer ## architecture - Distributed Cloud Run pipeline - GCS Event Triggers -> Transcribe -> Vectorize - Hybrid Vector + Semantic Search with Reranker ## sophistication - Async state machine tracking - Graceful fallback chains (Vertex Speech -> Whisper) - Hash-based deduplication ## stack - Google Cloud Run - Vertex AI - Firebase - Next.js --- # Guestlists Miami VIP nightlife management system with a public-facing AI concierge, enforcing strict write-only security models. Source: https://www.iamfern.com/projects/guestlists-miami Role: Full Stack Developer ## architecture - Dual-agent system (Public 'Baddie' + Admin) - Write-only security model for public agent - Event-driven Firestore triggers ## scale - API rate limiting (20 req/min) - 21,904 lines of API code - Multi-tenant architecture with 3-tier hierarchy ## stack - Firebase - Gemini Flash - React - Node.js --- # Reddit Email Digest LangGraph pipeline synthesizing daily and weekly Reddit newsletters using dual-model abstraction and structured parsing. Source: https://www.iamfern.com/projects/reddit-digest Role: Developer ## architecture - 10-agent pipeline (Daily vs Weekly modes) - Dual-model abstraction (GPT-4o / DeepSeek) - Pydantic JSON schema enforcement ## Results & scale - 33.9% character reduction - 32.6% token reduction via semantic dedup ## stack - LangGraph - DeepSeek - Pydantic - Python --- # Auto Job Applier Resilient Selenium-based bot with pluggable AI adapters (DeepSeek, OpenAI, Gemini) for automated job applications. Source: https://www.iamfern.com/projects/auto-job-applier Role: Developer ## architecture - Multi-AI provider adapter pattern - Three-tier fallback strategy (AI -> Rule -> Fallback) - Session state preservation ## scale - 100+ applications/hour capability - Handles 8 dynamic question types ## stack - Selenium - Python - DeepSeek - Gemini --- # Project DREA Privacy-first emotional intelligence platform using Vertex AI RAG to analyze relationship dynamics and detect manipulation patterns. Source: https://www.iamfern.com/projects/project-drea Role: Lead Engineer ## architecture - Vertex AI Discovery Engine (5 pillar datastores) - Multi-modal input (Text, Voice/Whisper, OCR) - Async Cloud Functions pipeline ## sophistication - Manipulation pattern detection (Gaslighting, DARVO) - PII Redaction & AES-256 Encryption ## stack - Vertex AI - Firebase - TypeScript - Google Cloud --- # Pocket Officer Real-time Florida statute violation identifier using Vertex AI Search grounding and structured output validation. Source: https://www.iamfern.com/projects/pocket-officer Role: Developer ## architecture - Gemini 2.5 Flash with Vertex AI Search grounding - Multi-pass enrichment pipeline - Vercel Serverless Python WSGI ## stack - Vertex AI Search - Next.js - Python - Stripe --- # Semantic Deduplication CPU-optimized semantic clustering engine achieving ~34% token reduction for RAG pipelines via SimHash and graph clustering. Source: https://www.iamfern.com/projects/semantic-dedup Role: R&D ## architecture - SimHash near-duplicate prefiltering - NetworkX graph clustering - ChromaDB persistence ## Results & scale - 33.9% character reduction avg - Fast CPU-based processing ## stack - Python - NetworkX - ChromaDB - SentenceTransformers --- # YouTube Deep Dive Precision analysis tool for technical presentations, extracting deep architectural patterns from video transcripts. Source: https://www.iamfern.com/projects/youtube-learning Role: Developer ## architecture - Full transcript single-pass processing - Dual-temperature strategy (0.0 analysis / 0.1 gen) - YouTube Transcript API ## stack - LangChain - DeepSeek - Python --- # Raggity Chat POC Workspace utilizing OpenAI Assistants API with file_search and Google Drive integration for collaborative RAG. Source: https://www.iamfern.com/projects/raggity-chat Role: Developer ## architecture - OpenAI Assistants API + Vector Store - Dual parallel pipelines (Upload -> Index + Drive) - Thread-based conversation persistence ## stack - React - Express - OpenAI API - Google Drive API --- # CourtListener MCP Specialized MCP server for querying legal opinions and filings via the CourtListener API. Source: https://www.iamfern.com/projects/court-listener-mcp Role: Author ## architecture - MCP Protocol binding - 7 specialized legal research tools ## stack - MCP - Python - CourtListener API --- # Agentic AI & Delivery Build agents that investigate issues, use tools, and prepare tested changes for review. Source: https://www.iamfern.com/services/genai-strategy ## Focus - Multi-agent architecture - Autonomous development workflows - Evaluation and release controls ## Approach Start with one workflow and a clear acceptance test. Define tool access, failure handling, and the human review point before expanding autonomy. ## Related projects - [Autonomous Software Delivery](https://www.iamfern.com/projects/autonomous-sdlc) - [Security Review Agents](https://www.iamfern.com/projects/security-review-agents) - [Meeting-to-Backlog Planner](https://www.iamfern.com/projects/meeting-to-backlog) --- # AI Strategy & Cost Compare the options. Measure quality and cost. Build the business case before the commitment. Source: https://www.iamfern.com/services/ai-tool-evaluation ## Focus - Build-vs-buy evaluation - Provider benchmarking - AI cost and value modeling ## Approach Compare providers against the same tasks and evidence. Separate measured spend from projected savings, and include integration and operating costs in the decision. ## Related projects - [AI Product Scorecard](https://www.iamfern.com/projects/ai-leadership-dashboard) - [Enterprise AI Assistant](https://www.iamfern.com/projects/enterprise-ai-assistant) - [Semantic Deduplication](https://www.iamfern.com/projects/semantic-dedup) --- # Enterprise Knowledge & ServiceNow Connect company knowledge to useful answers, with scoped access and traceable sources. Source: https://www.iamfern.com/services/enterprise-ai-innovation ## Focus - RAG and ingestion pipelines - ServiceNow architecture - MCP tools and integrations ## Approach Organize sources around the questions people actually ask. Make ingestion repeatable, check access at retrieval, and keep source references attached to the answer. ## Related projects - [Enterprise Knowledge Platform](https://www.iamfern.com/projects/enterprise-knowledge-platform) - [ServiceNow Service Modeling](https://www.iamfern.com/projects/csdm-inference) - [Enterprise AI Assistant](https://www.iamfern.com/projects/enterprise-ai-assistant)