Work
Projects
Things I'm building across agents, scientific search, code intelligence, and evaluation. For how they connect, try the interactive graph.
-
An orchestration-builder SDK for multi-agent coding workflows. I'm a maintainer.
GoAgentsOrchestration
-
A benchmark for evaluating how well coding agents understand and navigate code across large, distributed enterprise codebases.
PythonEvaluationAgents
-
Keep docs in sync with code. Livedocs extracts structural claims from source into per-repo SQLite databases that AI agents query over MCP, no expensive grep-and-read cycles.
GoMCPtree-sitter
-
A benchmark suite for evaluating how AI coding agents use external context-retrieval tools on realistic developer tasks in large, enterprise-scale codebases.
PythonEvaluationRetrievalDocker
-
A theory-of-mind agent for Claude Code that learns your coding preferences, interaction style, and project patterns across sessions.
TypeScriptAgentsMemory
-
Aggregates feeds and presents curated weekly and monthly digests of code intelligence, tools, and AI agents using hybrid LLM + BM25 + recency scoring.
TypeScriptRetrievalLLM
-
A background agent that checks organization-wide code invariants across every repository indexed by Sourcegraph, triggered by PR events and a weekly cron.
TypeScriptAgentsMCP
-
A behavioral taxonomy, annotation framework, and shareable dataset backend for analyzing why coding agents succeed or fail on benchmark tasks.
PythonEvaluationAgents
-
An agentic research assistant over the NASA SciX / ADS corpus, bridging AI agents with scholarly search infrastructure.
PythonAgentsMCPRetrieval
-
Coding standards, agent roles, skills, and multi-step workflows that read the same whether you drive Claude Code, Codex, Amp, or anything that reads an AGENTS.md.
JavaScriptAgents
-
A tiered-oracle funnel for evaluating automated code migrations end to end, Java 8 to 17, Python 2 to 3, with a pluggable ecosystem.
PythonEvaluationMigration
-
An MCP tool agentic-experience evaluation framework, measuring how usable MCP tools actually are for agents.
PythonEvaluationMCP
-
Measuring how LLM-powered tools discover, recommend, and describe products. GEO is the AI equivalent of SEO.
EvaluationRetrievalLLM
-
Benchmarks AI coding agents against your own codebase by mining evaluation tasks from its git history, so the suite can't be contaminated by training data.
PythonEvaluationAgents
-
A browser co-op deck-builder in the Slay-the-Spire / Ascension lineage, rendered entirely in hand-authored stained-glass art and tuned to be readable by a six-year-old.
ReactZustandViteTypeScript
-
Fine-tuning infrastructure for converting natural language into ADS / SciX scientific-literature search queries.
PythonNLPSciX
-
Reusable packs for Gas City. The PR-pipeline and Slack packs are mine.
GoPythonAgents
-
A dashboard for Gas City multi-agent orchestrations.
TypeScriptReact
-
A booth game for the AI World's Fair. Guess how much code AI agents are writing on GitHub, and a wave of agent-written code crashes in.
NodeExpressPostgres
-
A Wheel of Fortune practice app built with React Native and Expo, three game modes, real puzzle packs, seeded for repeatable practice.
React NativeExpoTypeScript
-
Build and benchmark agentic memory using a multi-agent orchestrator's own work traces as the evaluation corpus, where every unit of work carries a real lifecycle outcome and a full trace.
TypeScriptPythonEvaluationAgent memory
-
A stateless MCP server on Cloud Run that gives any authenticated Sourcegraph employee, through claude.ai, one tool surface over curated per-account research (GCS corpus) and live internal data (Salesforce, Looker, PostHog, HubSpot via cost-safeguarded databot), spanning account discovery, intelligence, lead scoring, and voice-checked outreach drafting.
PythonAgentsSlackLLM
-
sjarmak.ai: an Astro static site whose content collections form a typed knowledge graph. This very project entry is one node in it.
AstroTypeScriptZodCytoscapeMDX
-
Measures whether a repository actually works for AI coding agents by running an agent against tasks mined from its own git history and scoring what it did, instead of checking for the presence of files like AGENTS.md.
RustEvaluationAgents