Eric Provencher

Activities

Engineer with background in XR and software development; publicly referenced as an XR engineer at Unity in multiple profiles and interviews (sources listed below).
Founder and creator of RepoPrompt, a native macOS “context engineering” toolbox for AI-assisted development. Developed RepoPrompt to solve practical problems that arise when feeding large language models (LLMs) with project context.
Early work and public commentary focus on how to provide precise, token-efficient context to models and on avoiding “context rot” when using AI for real-world codebases.

Active maintainer and product lead for RepoPrompt (desktop macOS app and MCP server mode). Primary focus: improving context-building tooling, multi-repo reasoning, and model integrations (e.g., Claude, Gemini, other LLMs).
Advocates a “human-in-the-loop” approach to tool-calls: combine curated context with model calls and review before applying changes.
Publishes writings and demos on context engineering best-practices and multi-model workflows.

Founder of RepoPrompt (product/company association).
Professional connections to the wider AI-assistant ecosystem; notable integrations and interoperability with tools like Claude Desktop and agent-based workflows (via the Model Context Protocol / MCP).
Employment / prior role references point to Unity (XR engineering).

Continued evolution of context engineering tools: expect improvements in automatic context builders (codemaps, token-aware selection) and tighter integrations with agent frameworks.
Growth of “MCP-like” interoperability: RepoPrompt’s MCP server mode signals a trend toward standardized context-sharing between desktop apps and AI agents.
Emphasis on multi-model workflows (driver/reasoner split) where lighter orchestration models manage flow and stronger reasoning models handle complex analysis and code-writing tasks.

Context engineering, developer tooling, reliable AI-assisted refactors and tests, and developer workflows that balance human oversight with AI automation.
Improving token efficiency and practical model selection strategies for real-world codebases.

Context curation: Use RepoPrompt to visually browse a repository, create “presets” of files or folders relevant to a task (e.g., a service + its tests), then send that curated context to an LLM for focused refactors or test generation.
Multi-repo reasoning: Load multiple repositories in a workspace when working on cross-repo changes (shared libraries, infra changes) so the model understands dependencies and references across projects.
MCP integration: Run RepoPrompt in MCP server mode to let agent workflows programmatically request context snapshots, build prompts, and (with human oversight) apply suggested changes.

PerplexityAI web-searches and syntheses on “Eric Provencher” and “RepoPrompt” (context engineering, MCP integration, product features and updates).
Public interviews, product notes and demo write-ups discussing RepoPrompt’s Context Builder, code maps, and multi-model workflow practices.