project / Aug 2025 – Present

Lattice

Local agentic router that runs multi-step builds and tests in isolated workspaces, records structured logs, and packages only fresh deliverables.

view github reporesume
timelineAug 2025 - Present
focusai
stack
  • Python
  • Groq
  • Gemini
  • LM Studio

problem

AI-assisted build and test runs are hard to keep isolated, auditable, and easy to hand off when outputs mix with the source tree and logs are scattered.

approach

A Python CLI routes work across built-in or configurable subagents, gives each run its own seeded workspace and artifact directory, and exposes run inspection through logs or an optional FastAPI server.

implementation notes

  • Packaged a Python 3.9+ CLI exposed as `lattice` with run, logs, scrub, config, and serve commands.
  • Created isolated per-run directories containing run.jsonl, workspace files, artifacts, transcripts, summaries, and deliverables.
  • Seeded workspaces from the current directory by default, with an empty-workspace mode through LATTICE_WORKSPACE_SEED=empty.
  • Organized router/subagent modules for backend, frontend, LLM API, tests, toolbox variants, dynamic Python tools, contracts, stage gates, and RAG-related workflows.
  • Added test coverage across APIs, providers, artifacts, contracts, concurrency, command validation, stage gates, subagent libraries, and router regressions.

impact

  • Isolates generated work from the source tree
  • Preserves auditable run logs and transcripts
  • Packages only files created or modified during a run into a filtered deliverable archive

links