Video Agent Context Graph ↗
Repository under Adams’s verified GitHub account, also linked from his personal site. Combines TwelveLabs video understanding, OpenAI structured extraction and entity canonicalization, AWS Strands orchestration, and Neo4j vector retrieval and graph exploration. Videos connect to ordered segments, entities, and topics; normalized entity and topic names connect material across recordings. The README identifies create-context-graph as its starting point and describes community maintenance rather than official support.
github.com
AI Iceberg deep-research demo ↗
Collaborative project involving Temporal, Neo4j, Auth0, and Redpanda. Adams implemented the Neo4j memory layer using the Python driver and Cypher, storing research results, conversation history, and traces with vector-similarity retrieval and references to Temporal workflows for session recovery. His attributed contribution is memory and persistence; the source credits other contributors with orchestration, authentication, and event streaming.
github.com · Source ↗
Diving into the AI iceberg: What lies beneath your AI agents ↗
June 10, 2026 collaborative article coauthored with Melissa Herrera, Fred Patton, Peter Corless, and Chandler Mayo. Adams’s explicitly attributed section explains shared research memory, semantic retrieval, inspectable provenance, and recovery through stored workflow references. Traversal after vector retrieval and earlier interview termination are proposed extensions, not reported implemented behavior.
temporal.io
Fast Match Finder ↗
Adams explicitly describes building this graph-backed card-matching game in his Neo4j article. Humans and language models propose the shared symbol between two cards; a Cypher query over the card-and-symbol graph determines whether the answer is correct.
github.com · Source ↗
Dobble graph construction script ↗
Cypher script Adams identifies as his original Dobble graph script. Constructs a PG(2,7)-based deck with 57 cards and 57 symbols, connects each card to eight symbols, and verifies the deck invariants. The generated deck satisfies the mathematical constraints without necessarily reproducing the symbol combinations in his physical commercial deck.
gist.github.com · Source ↗
The math and the graph behind a popular match game ↗
Article authored by Adams on February 12, 2026. Explains finite projective planes through a card-matching game, models cards and symbols in Neo4j, verifies structural constraints, and introduces his Graph as Judge example for deterministic checking of human or model answers.
neo4j.com
Container Use for Locally Sandboxed, Background Agents in Zed ↗
Article authored by Adams on July 30, 2025. Explains how Container Use combines Dagger containers and Git worktrees for parallel background coding agents, including progress inspection, service tunnels, and terminal debugging. Recommends separate foreground and background tool profiles in Zed. Article authorship does not establish sole authorship of Container Use.
zed.dev
Trivy scanning module featured in The Daggerverse in Practice ↗
Dagger’s March 13, 2024 guide demonstrates a Trivy scan-image function from the jpadams/daggerverse/trivy module under Adams’s verified account. It packages container vulnerability scanning as a reusable Dagger function. The guide provides evidence of the account-hosted module, but its retained text does not identify Adams as the article’s author.
dagger.io

Bio, Work & Ideas

Jeremy Adams

Conference affiliation: Neo4j

On this page

Jeremy Adams is a senior developer advocate at Neo4j, helping developers build applications with graph databases, knowledge graphs, and AI. His career spans engineering, developer relations, and technical partnerships, with practical work ranging from security checks in build pipelines to shared memory for research agents.

From infrastructure automation to developer ecosystems

Adams worked in customer-facing technical roles at Sun Microsystems and Puppet before joining Twistlock, where his responsibilities included solutions architecture, business development, and alliances. He helped connect container security with developers’ existing workflows. In a 2019 implementation guide, he demonstrated how to scan container images within CircleCI and block a build from pushing an image to a registry if it failed vulnerability or compliance requirements. Developers could address problems while working on a feature, with security teams’ requirements applied automatically during the build.

At GitHub, Adams worked across developer relations and business development, including advocacy for GitHub Actions, its CI/CD automation platform. He also helped launch GitHub Enterprise purchasing through AWS Marketplace and announced the offering in December 2019. The arrangement addressed a practical adoption problem: customers wanted consolidated billing and purchasing compatible with their AWS enterprise agreements. The Marketplace offering ended in May 2021.

He later led ecosystem and developer-relations initiatives at Dagger, building community programs, partnerships, technical content, and integrations around its programmable CI/CD platform. His contributions included a Trivy container-scanning module featured in Dagger’s practical guide to the Daggerverse, an index of reusable functions. The module packaged vulnerability scanning behind a function developers could call from the command line or combine with other build and deployment steps.

His writing also extended into AI coding workflows. In a July 2025 guide to Container Use in Zed, Adams explained how containers provide isolated environments while Git worktrees give background agents separate working histories. He recommended separate tool profiles for foreground editing and background agent work, letting a developer continue editing while agents pursued parallel tasks without colliding with those changes.

Graphs for agent memory, video context, and reliable answers

At Neo4j, Adams builds examples that make connected data useful—and inspectable—within AI applications. His projects explore several distinct roles for graphs:

Shared research memory. In a deep-research application built with Temporal, Neo4j, Auth0, and Redpanda, Adams implemented the Neo4j memory layer. It stores research results, conversation history, and traces of how the work was produced. Vector similarity retrieves earlier research for semantically similar questions, allowing one user’s work to benefit others. The stored traces make that memory open to investigation; references to Temporal workflows also let the interface reconnect to a session after a browser reload. Adams described graph traversal after vector retrieval as a possible extension of the implementation.

Connections across videos. His Video Agent Context Graph project combines video understanding, structured extraction, vector search, and graph traversal. It organizes recordings into connected segments, entities, and topics, so an agent can retrieve relevant moments and explore related material across videos. Entities and topics use normalized names as keys: a shared subject can become a connecting node between recordings rather than a separate entry in each video.

Deterministic judging. In his graph-backed card-matching game, Adams represents cards and symbols as nodes, with relationships recording which symbols appear on each card. A Cypher query finds the symbol shared by two cards and checks a human or language model’s answer. He calls this Graph as Judge: for a fully specified game whose rules have been verified, the application can determine the answer by following relationships. The example gives developers a concrete way to distinguish a model’s proposed answer from a result conventional software can check.