Capabilities → evidence
Every claim links to an artifact: live demo, public code, eval report, or case study. No artifact, no claim.
Core
Shipped production agentic/LLM systems (8)
- Live grounded RAG chatbot — education client demo A live demo
Deployed a grounded RAG chatbot that passed a 33/33 golden-QA gate before ship.
- Live grounded RAG chatbot — education client demo B live demo
Second live deployment of the same pipeline, per-client tuned.
- Agora — feedback-analysis product demo live demo
Product case-study site with an animated real-trace pipeline stepper and a runnable Docker demo.
- LLM batch reliability (token-aware batching, integrity checks, staged retries) code exhibit
Production reliability engineering for high-volume LLM batch processing.
- Workflow dispatcher with human gates code exhibit
Staged agentic orchestration with pause semantics for human approval gates.
- Case study: Stoa — sovereign consultation analysis case study
Paid co-development of a sovereign AI platform, pilot-stage on real consultation data.
- Case study: Agora — feedback analysis product case study
Productised the platform pattern; accuracy benchmarked on SemEval-2014 gold data.
- Case study: FridaiOS — a personal agent platform case study
File-system-first agent platform — the same shape as an organisational agent deployment.
RAG end-to-end (4)
- Live grounded RAG chatbot — education client demo A live demo
Deployed a grounded RAG chatbot that passed a 33/33 golden-QA gate before ship.
- Hybrid RAG retrieval (RRF) on Cloudflare Workers code exhibit
Keyword+vector retrieval fused by RRF with graceful degradation, in production demos.
- Case study: Stoa — sovereign consultation analysis case study
Paid co-development of a sovereign AI platform, pilot-stage on real consultation data.
- Case study: website→demo delivery pipeline case study
7 gated stages from prospect URL to live grounded chatbot + priced proposal.
Eval harnesses (5)
- Live grounded RAG chatbot — education client demo A live demo
Deployed a grounded RAG chatbot that passed a 33/33 golden-QA gate before ship.
- Golden-QA eval loop (evals as ship gates) code exhibit
Chatbots must pass the full golden suite before and after deploy.
- LLM-judge eval report + cost tracker eval report
Model selection driven by explicit quality/cost/latency pass-fail gates.
- Document generation pipeline (skills-as-code) repo
Facts-grounded document generation with keyword-coverage checks and human checkpoints.
- Case study: Agora — feedback analysis product case study
Productised the platform pattern; accuracy benchmarked on SemEval-2014 gold data.
Observability (2)
- Multi-provider model router + token/cost ledger code exhibit
Ordered fallback chains with schema-validated retries and per-call cost accounting.
- Case study: FridaiOS — a personal agent platform case study
File-system-first agent platform — the same shape as an organisational agent deployment.
Cost/performance engineering (5)
- Agora — feedback-analysis product demo live demo
Product case-study site with an animated real-trace pipeline stepper and a runnable Docker demo.
- LLM batch reliability (token-aware batching, integrity checks, staged retries) code exhibit
Production reliability engineering for high-volume LLM batch processing.
- Multi-provider model router + token/cost ledger code exhibit
Ordered fallback chains with schema-validated retries and per-call cost accounting.
- LLM-judge eval report + cost tracker eval report
Model selection driven by explicit quality/cost/latency pass-fail gates.
- Model pricing analysis (cost engineering) brief
Derived per-job cost from token mechanics; caught a 6–7× pricing error in code constants.
Structured outputs + schema validation (4)
- LLM batch reliability (token-aware batching, integrity checks, staged retries) code exhibit
Production reliability engineering for high-volume LLM batch processing.
- AWS Bedrock + LangChain structured outputs code exhibit
Cloud LLM integration via langchain_aws with schema-validated structured outputs.
- Multi-provider model router + token/cost ledger code exhibit
Ordered fallback chains with schema-validated retries and per-call cost accounting.
- Outreach engine (schema-validated, proof-gated copy) repo
Personalized outreach bounded to proof-gated claims — an anti-fabrication guardrail.
Docker + CI/CD + git hygiene (3)
- agora-demo — runnable Docker bundle repo
Clone → docker compose up → logged-in analysis dashboard in ~5 minutes.
- Per-engagement VPS deploy system (Caddy + Docker) code exhibit
Multi-tenant isolation with auto-TLS and per-engagement generated secrets.
- Docker image leak scanner code exhibit
Proves proprietary code is absent from every published image layer.
Production Python (2)
- LLM batch reliability (token-aware batching, integrity checks, staged retries) code exhibit
Production reliability engineering for high-volume LLM batch processing.
- MCP server (FastMCP, Streamable HTTP) code exhibit
Hand-written MCP server — 5 tools, 3 resources, 3 prompts, auth, per-tool rate limits.
Business translation (3)
- Document generation pipeline (skills-as-code) repo
Facts-grounded document generation with keyword-coverage checks and human checkpoints.
- Outreach engine (schema-validated, proof-gated copy) repo
Personalized outreach bounded to proof-gated claims — an anti-fabrication guardrail.
- Case study: Stoa — sovereign consultation analysis case study
Paid co-development of a sovereign AI platform, pilot-stage on real consultation data.
End-to-end lifecycle ownership (2)
- Document generation pipeline (skills-as-code) repo
Facts-grounded document generation with keyword-coverage checks and human checkpoints.
- Case study: website→demo delivery pipeline case study
7 gated stages from prospect URL to live grounded chatbot + priced proposal.
Frequently required
LangChain / LangGraph / named frameworks (3)
- AWS Bedrock + LangChain structured outputs code exhibit
Cloud LLM integration via langchain_aws with schema-validated structured outputs.
- Workflow dispatcher with human gates code exhibit
Staged agentic orchestration with pause semantics for human approval gates.
- Harness ↔ LangGraph equivalence map diagram
Concept-for-concept mapping from my orchestration engine to LangGraph primitives.
Cloud AI platforms (4)
- AWS Bedrock + LangChain structured outputs code exhibit
Cloud LLM integration via langchain_aws with schema-validated structured outputs.
- Model pricing analysis (cost engineering) brief
Derived per-job cost from token mechanics; caught a 6–7× pricing error in code constants.
- Terraform (GCP Cloud Run, Pub/Sub, IAM) code exhibit
Authored IaC for a production-shaped product; separately operated an upstream AWS stack.
- Cloud portability brief (AWS ↔ Azure) brief
What I ran on AWS, mapped service-for-service to the Azure AI stack.
Workflow automation tools (2)
- Case study: FridaiOS — a personal agent platform case study
File-system-first agent platform — the same shape as an organisational agent deployment.
- Sales-ops automation architecture (schematic) diagram
The n8n/Gmail→HubSpot automation stack I built in a sales role — rebuilt as a schematic.
API / MCP / integration breadth (4)
- MCP server (FastMCP, Streamable HTTP) code exhibit
Hand-written MCP server — 5 tools, 3 resources, 3 prompts, auth, per-tool rate limits.
- Outreach engine (schema-validated, proof-gated copy) repo
Personalized outreach bounded to proof-gated claims — an anti-fabrication guardrail.
- Case study: FridaiOS — a personal agent platform case study
File-system-first agent platform — the same shape as an organisational agent deployment.
- Sales-ops automation architecture (schematic) diagram
The n8n/Gmail→HubSpot automation stack I built in a sales role — rebuilt as a schematic.
Postgres / pgvector (2)
- Hybrid RAG retrieval (RRF) on Cloudflare Workers code exhibit
Keyword+vector retrieval fused by RRF with graceful degradation, in production demos.
- Case study: Stoa — sovereign consultation analysis case study
Paid co-development of a sovereign AI platform, pilot-stage on real consultation data.
Human-in-the-loop + guardrails (4)
- Agora — feedback-analysis product demo live demo
Product case-study site with an animated real-trace pipeline stepper and a runnable Docker demo.
- Workflow dispatcher with human gates code exhibit
Staged agentic orchestration with pause semantics for human approval gates.
- Case study: Stoa — sovereign consultation analysis case study
Paid co-development of a sovereign AI platform, pilot-stage on real consultation data.
- Case study: Agora — feedback analysis product case study
Productised the platform pattern; accuracy benchmarked on SemEval-2014 gold data.
Data governance / sovereignty (3)
- Per-engagement VPS deploy system (Caddy + Docker) code exhibit
Multi-tenant isolation with auto-TLS and per-engagement generated secrets.
- Docker image leak scanner code exhibit
Proves proprietary code is absent from every published image layer.
- Case study: Stoa — sovereign consultation analysis case study
Paid co-development of a sovereign AI platform, pilot-stage on real consultation data.
Conversational AI (5)
- Live grounded RAG chatbot — education client demo A live demo
Deployed a grounded RAG chatbot that passed a 33/33 golden-QA gate before ship.
- Live grounded RAG chatbot — education client demo B live demo
Second live deployment of the same pipeline, per-client tuned.
- Agora — feedback-analysis product demo live demo
Product case-study site with an animated real-trace pipeline stepper and a runnable Docker demo.
- Case study: Agora — feedback analysis product case study
Productised the platform pattern; accuracy benchmarked on SemEval-2014 gold data.
- Case study: website→demo delivery pipeline case study
7 gated stages from prospect URL to live grounded chatbot + priced proposal.
Discovery / pilot / POC craft (2)
- Case study: Stoa — sovereign consultation analysis case study
Paid co-development of a sovereign AI platform, pilot-stage on real consultation data.
- Case study: website→demo delivery pipeline case study
7 gated stages from prospect URL to live grounded chatbot + priced proposal.
Terraform / IaC (2)
- Terraform (GCP Cloud Run, Pub/Sub, IAM) code exhibit
Authored IaC for a production-shaped product; separately operated an upstream AWS stack.
- Cloud portability brief (AWS ↔ Azure) brief
What I ran on AWS, mapped service-for-service to the Azure AI stack.
Context
AI-assisted engineering (Claude Code) (1)
- Case study: FridaiOS — a personal agent platform case study
File-system-first agent platform — the same shape as an organisational agent deployment.