Kevin Lewis

Projects

Applied projects and practice-based case studies focused on cloud platforms, responsible GenAI systems, agentic workflows, and evidence-based technical decision-making.

View ExperienceAsk the Assistant

Applied Projects

Concrete builds with real constraints: cloud delivery, AI-assisted workflows, retrieval, evaluation, cost controls, and human-in-the-loop design.

GenAI Portfolio Assistant

Applied

An evidence-based assistant embedded in this site that answers recruiter questions using curated knowledge, retrieval-style patterns, and guardrails against unsupported claims.

GenAIRAGGuardrailsOpenAI APINext.js
  • Evidence-grounded recruiter Q&A
  • Refusal and redirect patterns for unsupported claims
  • Cost-aware OpenAI API integration

Status: Active

OtaskuHub

Applied

A central AI operations and portfolio-control dashboard for app registry data, telemetry, agent activity, prompt governance, LLM usage, cost visibility, and human review workflows.

AI OperationsTelemetryPrompt GovernanceAgent RunsLLM Cost TrackingSupabaseVercel
  • App registry and API-key integration model
  • Agent run/step tracking with metadata and failure visibility
  • Prompt versions, pending approvals, LLM cost summaries, and app health checks

Status: Active / Building

OtaskuHub MCP Bridge

Applied

A local Model Context Protocol bridge that wraps selected OtaskuHub endpoints as read-only tools for MCP-capable AI clients such as Codex Desktop and Claude Code.

MCPCodexClaude CodeNode.jsTypeScriptTool WrappingAPI-Key Integration
  • Local stdio MCP server with built dist/server.js entrypoint
  • Read-only wrappers for app health, cost summaries, agent runs, prompt versions, and approvals
  • Hub remains the source of truth while the bridge limits direct data access

Status: Active / Building

OtaskuScout

Applied

A full-stack agentic job-search platform that ingests Gmail job alerts and manual postings, deduplicates opportunities, evaluates role fit, and generates human-approved recommendations.

Agentic AIHuman-in-the-loopDeep Fit AnalysisLangGraphLLM EvaluationToken/Cost TrackingNext.jsPostgreSQLRedis/CachingOpenAI APIGmail IngestionDecision Learning
  • Persistent agent runs with status, summaries, timing, token usage, and estimated cost
  • LangGraph-based workflow patterns for agentic task flow
  • Task proposals with pending, approved, dismissed, and completed outcomes

Status: Active / Building

OtaskuLang

Applied

A mobile-first AI language practice app that generates a daily Bento of personalized learning cards for Japanese, Spanish, and Korean.

AI Language LearningMobile AppExpoReact NativeOpenAI APISupabase AuthNeon PostgresRedis/CachingGenerated Content
  • AI-generated Daily Bento packs with vocabulary, grammar, story, quiz, speaking, listening, and practice cards
  • Language onboarding and level-based preferences for Japanese, Spanish, and Korean
  • Text-only AI conversation practice with corrections, vocabulary suggestions, and recap generation

Status: Active / Building

OtaskuAI

Applied

An exploratory anime recommendation project focused on similarity, ranking, retrieval, explainability, cost-aware architecture, and uncertainty-aware UX.

RecommendersEmbeddingsRetrieval/RankingExplainabilityCost-Aware
  • Similarity + ranking
  • Explainable recommendations
  • Cost-aware architecture

Status: Planned / Building

ParaAIS

Applied

An experimental AI interaction framework focused on trust boundaries, uncertainty, evidence-aware responses, helpful refusals, and safe redirection.

Trust BoundariesEvidence-Aware UXGuardrailsAI GovernanceSystem Design
  • Tiered claim confidence
  • Separation between retrieved knowledge and model reasoning
  • Helpful refusals and safe redirection

Status: Concept / Prototype

Case Studies

Independent, practice-based write-ups focused on constraints, decisions, and tradeoffs. These are based on small-scale projects or publicly available material—not employer-specific implementations.

Cost-Aware & Ethical AI Deployment (OtakuAI)

Case StudyIndependent

Practice-based case study examining how a deliberately over-provisioned, small-scale AI system (OtakuAI) was redesigned to demonstrate cost-aware architecture decisions and ethical AI governance using AWS tooling and formal risk frameworks.

AI ArchitectureCost AwarenessAWSEthical AINIST AI RMF
  • Modeled baseline → optimized design (projected ~90% reduction)
  • Rightsizing, lifecycle controls, and usage constraints
  • Lightweight monitoring aligned with NIST AI RMF principles

Status: Documented

AI Ethics & Governance Risk Assessment (Grok Case Study)

Case StudyIndependent

Practice-based AI governance case study applying formal risk frameworks to a real-world generative AI system, resulting in a documented risk register and mitigation recommendations.

AI GovernanceAI EthicsNIST AI RMFOECD PrinciplesRisk Analysis
  • Authored AI ethics risk memo analyzing xAI’s Grok chatbot
  • Applied NIST AI RMF and OECD Principles to structure risk assessment
  • Proposed mitigations: red-teaming, moderation buffers, transparency reporting

Status: Documented