All 23 Projects
ToolChainDev
Python · FastAPI · LangGraph · ChromaDB · Pydantic · DEPLOYED
Multi-agent RAG platform. Supervisor → RAG/Search/Explain agents.
Agents, RAG, vector search, Python, tool agents, production rigor
Paper Candle
Next.js · Workers AI · M3/Gemma/GLM · DEPLOYED
Market education. Multi-model router, SRS, 686 sources, AI resilience.
LLM eval, model selection, managed services, resilience, prompts
Asset Persona / Upgrade.self
Hono · Drizzle · Supabase · Workers AI · DOCKER
AI learning platform + RAG engine. Provenance, BM25, FSRS, assessments.
RAG, responsible AI, provenance, eval, managed services
InHouse
Laravel · Inertia · Elasticsearch · DOCKER
Regulated compliance marketplace. 43 models, LegalComplianceService.
Regulated env, compliance, audit trails, ES, Docker
GrazzHopper (v1)
Laravel · Inertia · Elasticsearch · GIT
Compliance platform w/ voice agent. ComplianceLaw, jurisdiction routing.
Regulated env, compliance, voice agent, multi-jurisdiction
SMS Marketing
Next.js · Prisma · PostgreSQL · DEPLOYED
TCPA-compliant campaigns. Quiet-hours, opt-out, consent audit trails.
Regulated env, compliance guardrails, audit trails
SocialStakes
React Native · Expo · Supabase · DEPLOYED
Mobile social gaming. ML interaction engine, real-time data.
Mobile, real-time data, ML engine
Skills Library v2
Markdown · Next.js · DEPLOYED
177 structured agent skills. 29 workflows. 4,500+ lines.
Prompt authoring, reusable components, frameworks, mentorship
Automation Centre
n8n · 758 FILES
Workflow orchestration. API integrations, webhooks, data pipelines.
Workflow orchestration, API integration, automation
gemini-cli
Node.js · Docker · GIT
Google Gemini CLI tool. Docker containerized.
Managed services (Gemini), CLI, Docker
video-watermark-remover
Python · FastAPI
Video processing tool.
Python, media processing
Agentic Study Hall
Next.js · GIT
Learning platform frontend. Assessments, spaced repetition.
Context engineering, learning systems
play4keeps-web
Next.js · GIT
Web app, full-stack.
Full-stack engineering
manga-scroll-feed
Vite · React · Firebase
Educational feed, interactive learning games.
Content delivery, interactive learning
storyboard
CF Worker · Vercel · DEPLOYED
Asset imagery and storyboard platform.
Full-stack, cloud deployment
maaricare-site
CF Worker · DEPLOYED
Healthcare brand site on Cloudflare.
Cloud deployment, regulated (healthcare)
Bag of Agents
Markdown docs
Agent strategy documentation.
Multi-agent design patterns
figma-web-app
Node.js · GIT
Figma integration tool.
API integration
hermes-agents
Scripts · configs
Agent automation scripts and configs.
Automation, scripting
workspace
Mixed source files
General workspace.
General engineering
clawd
Symlink → .openclaw
OpenClaw agent framework.
Agent frameworks, open-source
GrazzHopper v2
Next.js · Supabase · CONFIDENTIAL
Cannabis compliance platform (active dev). Eval, pgvector, DR runbook.
CONFIDENTIAL — talk patterns only, never show
What You'll Do (Day to Day)
8/9 covered
Design & ship mission-grade GenAI: agentic workflows, RAG, low hallucination, tight p95 latency, predictable cost
"Mission-grade" = reliable enough for government. p95 = 95th percentile response time.
COVERED
ToolChainDev ✓
Paper Candle ✓
Asset Persona △
Evidence: ToolChainDev has max 5 iterations, low temperature, grounding prompts, Prometheus metrics. Paper Candle has AI resilience (retry/backoff), token cost tracking.
"I build with guardrails: max iterations, low temperature, grounding prompts, human-in-the-loop for low-confidence outputs."
Agent frameworks & orchestration: LangChain/LlamaIndex/Semantic Kernel; task decomposition, tool use, guardrails, recovery/fallback
LangChain = general agents. LlamaIndex = RAG-focused. Semantic Kernel = Microsoft/Azure.
COVERED
ToolChainDev ✓
Paper Candle ✓
Automation △
Evidence: ToolChainDev: LangGraph supervisor → RAG/Search/Explain agents with Pydantic routing. Paper Candle: AI router (M3/Gemma/GLM), tool registry, retry with backoff, circuit breaker.
Platform integration (no model training): AWS Bedrock, Azure OpenAI, Google Vertex AI, Amazon Kendra, managed services (Document AI, Gemini, Gemma)
Use managed AI services. No training. Just call APIs.
GAP
Paper Candle ✓ (Gemma)
Asset Persona ✓ (Gemma)
Skills Lib △ (documented)
gemini-cli ✓
Have: Gemma on Workers AI. Gemini via gemini-cli. Documented Bedrock, Azure OpenAI, Vertex.
GAP: Not shipped on Bedrock or Vertex yet. Ship ToolChainDev to Bedrock, build regulation agent on Vertex. Pennies for Bedrock, free for Vertex ($300 credit).
LLM selection & evaluation: compare models for quality, safety, latency, cost; author/test prompts; observability; safe rollback/fallback
Pick the right model. Test prompts. Monitor. Roll back if broken.
COVERED
ToolChainDev ✓
Paper Candle ✓
Asset Persona ✓
Skills Lib ✓
Evidence: Paper Candle routes between 3 models by task type. SRS engine evaluates learning outcomes. ToolChainDev has Prometheus + Sentry. Asset Persona has Hake/ECE/FSRS evaluation. 177 structured prompts.
RAG done right: vector search (Pinecone/Weaviate/OpenSearch/pgvector/FAISS/Chroma); chunking, metadata, NDCG
Build RAG properly. Chunk documents, tag metadata, evaluate with IR metrics.
COVERED
ToolChainDev ✓ (ChromaDB)
Paper Candle ✓ (686 sources)
Asset Persona ✓ (BM25)
InHouse ✓ (ES)
GrazzHopper ✓ (ES)
Evidence: ChromaDB + OpenAI embeddings in ToolChainDev. 686 chunked markdown files in Paper Candle. BM25 + provenance in Asset Persona. Elasticsearch in InHouse + GrazzHopper.
Production rigor: metrics/logs/traces, A/B experiments, incident playbooks, safety guardrails
Instrument everything. Test variations. Have runbooks.
PARTIAL
ToolChainDev ✓
Paper Candle ✓
Evidence: ToolChainDev: Prometheus, structlog, Sentry. Paper Candle: AI resilience (retry/transient/auth classification).
"I instrument before I scale. Prometheus metrics on every agent. Structured logs. Sentry. A/B via champion-challenger. DR runbook for incidents."
SRE & FinOps for AI: SLIs/SLOs (quality/latency/safety/cost), on-call, postmortems, MTTR; meter usage, optimize token/spend
SLI = metric. SLO = target. MTTR = Mean Time To Recovery. FinOps = cost optimization.
PARTIAL
ToolChainDev ✓
Paper Candle ✓
Evidence: Per-agent metrics in ToolChainDev. Token cost tracking in Paper Candle.
"SLIs I track: faithfulness >95%, p95 <3s, guardrail violations <0.1%, cost per query <$0.05."
Reusable platform components: SDKs, CI/CD templates, Terraform/IaC modules, evaluation harnesses
Build things other teams can reuse.
PARTIAL
ToolChainDev ✓
Skills Lib ✓ (177 skills)
Automation ✓ (758 workflows)
Evidence: Skills Library = 177 reusable agent skills. Automation Centre = 758 workflow files. ToolChainDev = deployable patterns.
"My skills library IS reusable components — 177 structured skills that other teams can plug into their agents."
Operate in real-world constraints: hybrid, restricted, air-gapped; Zero Trust; audit-ready controls
Air-gapped = no internet. Zero Trust = every component authenticates. Audit-ready = logged everything.
GAP
InHouse ✓ (audit)
GrazzHopper ✓ (compliance)
SMS Mktg ✓ (TCPA)
Have: 3 years in regulated compliance. LegalComplianceService with audit trail. TCPA consent logs.
GAP: Know the vLLM on-prem pattern. Say "compliance work taught me audit-first thinking. For air-gapped: vLLM + local embeddings + pgvector."
You'll Thrive Here If You Have
5/5 covered
End-to-end ownership: integration → deployment → observability → incident response
You built it, shipped it, monitored it, fixed it.
COVERED
ToolChainDev ✓
Paper Candle ✓
Asset Persona ✓
InHouse ✓
GrazzHopper ✓
SMS Mktg ✓
Evidence: 6 systems built, deployed, and maintained end-to-end. 3 deployed live.
Hands-on with LLMs, transformer-based apps, and RAG in production
Transformer-based app = any app calling an LLM (GPT, Claude, Gemini, Gemma — all transformers).
COVERED
ToolChainDev ✓
Paper Candle ✓
Asset Persona ✓
gemini-cli ✓
4 projects with LLM calls in production.
ToolChainDev ✓ (FastAPI/LangGraph)
video-watermark ✓
FastAPI, LangGraph, Pydantic, structlog, async/await. Full backend in Python.
Vector search & retrieval and grounding AI in enterprise/mission data
ChromaDB, Elasticsearch, pgvector, Vectorize.
COVERED
ToolChainDev ✓ (ChromaDB)
Asset Persona ✓ (Vectorize/BM25)
InHouse ✓ (ES)
GrazzHopper ✓ (ES)
Nice to Have
6/9 covered, 3 gaps
Integration with cloud AI services or on-prem inference stacks
Inference stack = vLLM, TGI, Ollama (on-prem). Bedrock/Vertex (cloud).
GAP
Paper Candle ✓ (Workers AI)
Asset Persona ✓
gemini-cli ✓
GAP: Know vLLM serving Llama pattern. Ship to Bedrock/Vertex.
LLM evaluation, prompt authoring/testing, A/B experimentation, LLM-Ops
COVERED
ToolChainDev ✓
Paper Candle ✓ (SRS)
Asset Persona ✓ (Hake/ECE)
Skills Lib ✓ (177 prompts)
Responsible AI (privacy, security, bias, transparency, human-in-the-loop) and data governance
COVERED
ToolChainDev ✓
Paper Candle ✓
Asset Persona ✓
InHouse ✓
GrazzHopper ✓
SMS Mktg ✓
Privacy: LegalComplianceService (consent/PII). Transparency: provenance + citations. HITL: GH moderation/scan.ts. Security: tool-scoped agents.
Tool-using agents for API integration and external data access
COVERED
ToolChainDev ✓ (Tavily)
Paper Candle ✓ (registry)
Automation ✓ (n8n APIs)
Containerization (Docker, K8s, VMware) + scripting (Bash, PowerShell)
COVERED
Asset Persona ✓ (Colima/Docker)
InHouse ✓ (Dockerfile)
GrazzHopper ✓ (bash)
gemini-cli ✓ (Docker)
Docker/Colima running Supabase containers now. Hermes daily. Bash scripting.
Prior work in regulated/secure environments (ATO, STIGs, Zero Trust)
COVERED
InHouse ✓
GrazzHopper ✓
SMS Mktg ✓
"3 years in regulated compliance. Understand Zero Trust. ATO/STIGs: working familiarity, would partner with security teams."
NVIDIA AI Foundations, OpenAI ChatGPT, AI-assisted dev tools (Cursor, Windsurf, Claude)
NVIDIA: NeMo = training, TensorRT = inference speed, NIM = containers
GAP
Daily: Cursor, Claude, Codex, Hermes ✓
GAP: Know the NVIDIA terms. Haven't deployed on NVIDIA hardware. Say "familiar with the ecosystem."
Contributions to frameworks/open-source; mentorship of engineers
COVERED
Skills Lib ✓ (177 skills)
Microsoft MANCODE ✓ (1,153)
Vineyard + Onion Dev ✓
Clear communication with engineers, PMs, and security/compliance stakeholders
COVERED
InHouse ✓ (legal→engineering)
Microsoft ✓ (taught)
Onion Dev ✓
Consulting background. Translated regulatory requirements into code at PLK. Taught 1,153 at Microsoft.
Case Study Structure
Case Study 1: ToolChainDev (12-15 min)
Deep-dive · multi-agent RAG
Architecture diagram, code walkthrough, guardrails. Your #1 anchor.
Covers: Agent frameworks, RAG, vector search, Python, tool agents, production rigor, FinOps
Case Study 2: Paper Candle (10-12 min)
Multi-model routing, SRS evaluation, AI resilience
The "LLM selection" story. Model router, resilience layer, learning evaluation.
Covers: LLM evaluation, model selection, managed services, production resilience, prompt authoring
Case Study 3 (optional): InHouse (5-8 min)
Regulated compliance, abstracted
LegalComplianceService, audit trails, Elasticsearch. No cannabis branding.
Covers: Regulated environments, compliance, audit trails, containerization
Talk About (no demo)
Skills Library → prompt authoring, reusable components
SMS Marketing → TCPA compliance, audit trails
Asset Persona → second RAG engine, provenance
n8n Automation → workflow orchestration
Docker/Colima → containerization
Microsoft/Vineyard/Onion Dev → mentorship
The 3 Gaps
close before interview
Gap 1: Bedrock / Vertex
Not shipped on AWS Bedrock or Google Vertex AI yet.
Fix: Ship ToolChainDev to Bedrock (under $1). Build regulation agent on Vertex (free $300 credit).
Gap 2: On-prem inference
No vLLM or Ollama deployed. Only API-based inference.
Fix: Install Ollama (10 min: brew install ollama, ollama run gemma2). OR know the vLLM talking point cold.
Gap 3: NVIDIA AI Foundations
Familiarity only. Haven't deployed on NVIDIA hardware.
Fix: NeMo = training. TensorRT = inference speed. NIM = containers. Say "familiar with the ecosystem."