Ruslan Shulga Open to senior/staff AI roles
VP Engineering at JPMorgan Chase. The systems I lead serve several thousand employees a day. Multi-agent orchestration and hybrid retrieval, plus the MCP servers behind them.
Updated July 2026 · Press ⌘ K for commands
About
For the last four years I've led AI platform engineering at JPMorgan Chase. My team built the multi-agent orchestration layer that several thousand employees use daily. We also wrote the hybrid RAG pipeline behind internal search and the MCP servers that connect Claude to internal APIs.
Before JPMC I spent two years on React and Next.js work at Earth Designs and two years running IT operations at PPS Capital. Nine years of production engineering total, four of them shipping AI in a regulated enterprise.
My focus right now is the boring infrastructure under agentic AI: evaluation pipelines that catch regressions before users notice, agents that fail safely when the model misfires. The stuff that rarely gets demoed but always decides whether the system holds up in production. I'm looking for what comes next.
Selected Work
Production multi-agent workflows used by several thousand employees firm-wide. LangGraph orchestrates Claude Agent SDK and OpenAI Agents SDK. Each agent has persistent memory and calls internal tools. Irreversible actions pause for a human checkpoint.
Replaced an embedding-only RAG with a hybrid pipeline: dense plus sparse retrieval, cross-encoder re-ranking, indexed across Pinecone, Weaviate, and pgvector depending on the use case. The pieces that mattered weren't the models. They were the eval harness and the index strategy per domain.
Custom Model Context Protocol servers that connect Claude to internal APIs and databases. Around 8 product teams use the stack. Features that used to take a sprint now take a day or two.
An AI gateway on top of AWS Bedrock and Azure OpenAI, with direct Anthropic API access for newer Claude models. Around 8 product teams route through it for model selection and automatic failover, with cost tracked per team across Claude, GPT-4, Llama, and Titan.
Document and image processing pipelines on Claude Vision and GPT-4V, with Gemini added recently for cost reasons. Handles the first pass of compliance reviews that used to sit with analysts.
Five agents operate ruslanshulga.com in public. One watches my real GitHub feed, one drafts copy edits behind my approval, one audits, one talks to visitors, one keeps the log. Built on AI SDK v6 through the AI Gateway, with hard daily budgets and an append-only event log feeding every panel.
One watches my real GitHub feed. One drafts site edits I approve by hand. One audits, one talks to visitors, one keeps the log. Every entry is timestamped and real, down to the dollar meter on their model budget.
Step onto the BridgeContact
Add the MCP endpoint to Claude under Settings → Connectors, or drop it in your editor's mcp.json. Six read-only tools, no auth. The chat on this site runs the same toolset and shows you its calls. Nightly quality scores live at /evals.