Skip to content
View homayoun-safarpour's full-sized avatar
0 followers · 9 following

Achievements

Achievement: Pull Sharkx3Achievement: YOLOAchievement: Quickdraw

Achievements

Achievement: Pull Sharkx3Achievement: YOLOAchievement: Quickdraw

Block or report homayoun-safarpour

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
homayoun-safarpour/README.md

Homayoun Safarpour

Agent CI stays green while the run is truncated, the handoff is null, or the MCP surface is wide open.

I build fail-closed gates for agent / judge / RAG systems — deterministic exit codes, frozen anchors, no vibe scores.

Ireland · Trustworthy AI / agentic systems · hire-facing OSS

pip install -e ".[dev]"   # in any gate repo below
homi-gate check-completion examples/pass.json
homi-gate check-handoff examples/handoff_ok.json
homi-gate check-mcp-allowlist examples/mcp_ok.yaml

Clone a named project below — not this profile repo.

Gate stack (start here)

Repo Job Fail signal
homi-gate Completion · handoff · MCP allowlist exit 1
trace-gate Trajectory vs frozen baseline exit 2
judge-drift-sentinel Score moved: system or judge? exit 2
agent-constraint-auditor Declared-constraint decay in transcripts exit 2
agent-loop-engine State · gates · one action · journal bounded loop

Eval & hire labs

Repo Job
judge-reliability-kit Why a judge panel disagrees (κ)
agent-eval-workbench Scenario traces + detectors
rag-eval-service Frozen RAG hit@k / MRR gates
ai-eng-skill-range 56 skills / 24 graded katas
agent-loop-field-guide Loop Contract before you automate
judge-field-guide Link-checked judge-tool map
repro-ml-pipeline sklearn → MLflow → FastAPI + signature CI

How I ship (house style)

  1. Lead with the failure mode, not the stack
  2. Pass + fail demo in one screen — no API key
  3. Table the asserts (command · pass · fail)
  4. Non-goals + composition (beside frameworks, not instead)
  5. Loud falsifier — claim dies if examples don’t go red

Weekly field remix (public)

Every week I take one strong public pattern (Promptfoo / DeepEval / AgentEvals / ECT / Hamel-style judges) and ship a thin Homi remix: same house voice, CI-green, hire-signal labels. Series tag: field-remix.

Links

Python 3.10–3.12 CI. Claims backed by named tests. Quickstarts under ~30 minutes.

Popular repositories Loading

  1. agent-constraint-auditor agent-constraint-auditor Public

    Audit agent transcripts for declared-constraint decay. Exit 0 CLEAN, 2 DECAY.

    Python 1

  2. judge-reliability-kit judge-reliability-kit Public

    Diagnose why an LLM judge panel disagrees: fix the rubric, or fix the task.

    Python 1

  3. agent-loop-engine agent-loop-engine Public

    Deterministic agent loop: state file, quality gates, one bounded action, append-only journal.

    Python

  4. judge-drift-sentinel judge-drift-sentinel Public

    Attribute eval-score movement to the system or the LLM judge using a frozen human anchor set and kappa.

    Python 2

  5. homayoun-safarpour homayoun-safarpour Public

    judge-drift-sentinel · judge-reliability-kit · agent-loop-engine · trace-gate · ai-eng-skill-range

  6. trace-gate trace-gate Public

    Fail CI when agent trajectory scores regress against a frozen baseline (exit 0 or 2).

    Python