Joseph Bergin

JPMorgan Chase · Wealth Management · Plano, TX

Building the shared retrieval layer that Wealth Management's AI agents run on, and the agents themselves.

• Centralized retrieval infrastructure

  Designed and built a centralized OpenSearch vector store with a RESTful API layer in front of it, so Wealth Management's AI agents draw document context from one shared service instead of each standing up its own retrieval stack.

• A RAG agent serving 1,000+ advisors

  Independently designed, built, and deployed a RAG agent on Google's ADK framework that answers investment questions for over 1,000 financial advisors.

• Reindexing at nine figures

  Built a hybrid regex-plus-LLM batch scoring pipeline and fine-tuned the LLM classification prompts behind it to reindex hundreds of millions of OpenSearch documents — cutting cost, reducing latency, and improving retrieval precision.

• Technical leadership

  Provided technical leadership for a team of four engineers: reviewing designs, answering implementation questions, and unblocking the decisions that were holding work up.

• Ownership and handoff

  Owned all feature development and deployment decisions as SME for a second production RAG agent, and led the full knowledge transfer during a cross-team handoff.

Akoe · Remote

Co-founded an LLM-powered quality assurance platform for grading and reviewing customer service calls. Acquired.

• From zero to acquisition

  Co-founded Akoe, an LLM-powered quality assurance platform for grading and reviewing customer service calls, and grew it through to a company sale.

• The grading engine

  Built the core engine that scored agent performance against customizable, admin-defined benchmarks — replacing a manual call-review process with something that ran on every call.

• Insights, not just scores

  Designed automated summary and insights reporting that surfaced call-quality trends and turned them into actionable recommendations for both agents and their managers.

• Full-stack and the infrastructure under it

  Deployed the full-stack application across Next.js/TypeScript, Python, and AWS, and owned the AWS infrastructure powering the platform's LLM grading pipeline.

JPMorgan Chase · Debit Tech · Plano, TX

Near-real-time event ingestion and the Terraform foundation the team deployed on.

• Near-real-time ingestion

  Designed and implemented a scalable pipeline for consuming near-real-time events using AWS Lambda and AWS Glue, giving downstream analytics low-latency access to data that had previously arrived in batches.

• Infrastructure as code

  Provisioned and managed the team's AWS infrastructure with Terraform, establishing repeatable, version-controlled deployments.

JPMorgan Chase · Consumer Banking · Plano, TX

Backend service work and zero-downtime schema migrations in production.

• Reconciling third-party data

  Implemented Spring Boot REST endpoints that transform and reconcile data from third-party APIs into consistent internal contracts.

• Zero-downtime migrations

  Authored Liquibase migration scripts to manage schema changes on production SQL databases without taking them offline.