JWConsultancy
01 What I build

AI that actually works for your business.

End-to-end AI systems for finance, e-commerce, and regulated industries — built by an engineer who has shipped at Johnson & Johnson, MSD, and Argenx.

02 Proof

Built AI systems at

Johnson & Johnson·MSD·Argenx·Greenyard
03 How we work

Two ways to work together

Track A

Fractional AI Engineering

You need a senior AI engineer embedded in your team on a sprint basis, not a consultant who delivers a report. Jelino builds and ships production AI systems alongside your team, and hands over something that actually runs.

Learn more →
Track B

AI Solutions Build

You have a problem and you need it solved. Jelino takes it from scoping to a shipped system. No prototype, no proof of concept. You end up with working software and the documentation to run it.

Tell me what you're building →
04 Under the hood

How it works

Agent → Router → Judge

LangGraph Orchestration

LangGraph Orchestration FlowAn agent graph mapping an Agent node to a Router node, which conditionally delegates to a Judge node.AgentRouterJudge
15% regression alarm · example

LLM-as-Judge Evaluation

8.4Score
Quality score (1–10)15% regression alarm
7 tiers · 5 providers

Cost-Routing Waterfall

Tier 1
Tier 2
Tier 3
Tier 4
Tier 5
Tier 6
Tier 7
OpenAI · Anthropic · Google · DeepSeek · Moonshot
Query → Embed → Retrieve → Rerank → Generate

RAG Pipeline

RAG Pipeline FlowA flow mapping Query -> Embed -> pgvector (active) -> Rerank -> LLM.QueryEmbedpgvectorRerankLLM
05 Featured build

Featured build

Case study

Dyniq

Problem

Route many specialised agents reliably and cheaply across a large model portfolio.

Outcome

Cost-optimal multi-provider routing with automated quality regression detection.

· 9 LangGraph state machines· 7-tier cost-routing cascade across 5 providers· 107 agents mapped to optimal models· LLM-as-judge: 15% regression alarm
LangGraph · PydanticAI · pgvector · Langfuse · LiveKitRead the case study →

Ready to build?

Start with a free scoping call.