Hi im Josh and this is my portfolio page, welcome. Built as a living creative tool and project canvas.

Work

AI-Native Research Platform

Worked across product, UX, and engineering to shape the researcher experience end to end, from board and document workflows to AI agents, retrieval, tabular data, integrations, and platform infrastructure.

Start a conversation

Lapis

Jul 2025 - Present

Founding Engineer / Full-Stack + AI Systems

Built an AI-native research platform with equal focus on product clarity, user workflows, and deep technical systems.

Lapis is an AI-native project management platform for research teams, not just a search product. The work spanned most of the platform: multi-surface product development, NeuralCore agent orchestration, semantic retrieval, document and knowledge-base systems, import and upload pipelines, tabular dataset handling, enterprise integrations, authentication and permissions, and the day-to-day UX researchers use to run work. Built across Next.js, Node.js, TypeScript, Supabase/Postgres, Azure/OpenAI, pgvector-style retrieval flows, and a growing data pipeline that supports both unstructured documents and structured datasets.

Product thinkingUX / UI implementationAI platform engineeringFull-stack systems

Overview

Lapis

Worked across product, UX, and engineering to shape the researcher experience end to end, from board and document workflows to AI agents, retrieval, tabular data, integrations, and platform infrastructure.

Platform

AI-native PM for research teams

Core systems

Agents, search, docs, datasets

Product lens

UX, workflows, systems thinking

Team

Built with a 5-person founding team

Detail

Platform

Built Lapis as a full product platform for research teams: project workspaces, board experiences, result capture, document systems, onboarding flows, permissions, and the infrastructure tying it all together. A large part of the work was not just engineering features, but shaping how the product should feel, flow, and support real research work.

AI + Retrieval

Developed the NeuralCore architecture around classification, planning, retrieval, execution, and response streaming. That included semantic search, project-scoped retrieval, document/entity handlers, context management, evaluation harnesses, and AI-native product behavior instead of a simple chat wrapper.

Documents + Data

Built the document and ingestion layer researchers actually operate in: uploads, imports, review states, embeddings, file viewers, knowledge-base organization, and structured dataset support for CSV/XLSX-style analytical workflows and tabular querying. That work also required thoughtful UX for dense information, status feedback, and multi-step workflows.

Product + UX

Contributed heavily to the product layer itself: translating complex AI and data capabilities into interfaces researchers could actually navigate, trust, and use. That included workflow design, interaction decisions, UI implementation, and making technically heavy features feel more coherent inside the platform.

Integrations + Access

Implemented and supported production integrations across GitHub, Google Drive, OneDrive, and SharePoint-style enterprise workflows, alongside authentication, scoped permissions, org/project isolation, and the operational plumbing required to make the platform usable in real teams.

Outcomes

Built NeuralCore, a staged AI assistant pipeline with intent routing, scoped retrieval, streaming responses, eval tooling, and workspace-aware behavior.
Implemented semantic search and document-grounded retrieval across research files, notes, boards, and project context.
Built upload, import, review, and ingestion flows for documents and structured datasets, including tabular querying paths for CSV and spreadsheet workflows.
Designed and shipped product flows researchers actually work inside, balancing complex capabilities with usable interfaces, information architecture, and clearer task-oriented UX.
Shipped document management, knowledge-base organization, permissions, authentication, and enterprise integrations across the product surface.

Stack

Next.jsNode.jsTypeScriptSupabasePostgreSQLAzureOpenAISemantic RetrievalAI AgentsTabular QueryingDuckDB PipelinesVector SearchRedisOAuth2GitHubGoogle DriveOneDriveSharePoint
Back to workDiscuss a similar project