AI Workflow Reliability Audit
For a workflow that works most of the time but fails in the ugly 10%.
- workflow and failure-mode map
- retry / timeout / duplicate-risk review
- prioritized remediation plan
- test checklist and handoff notes
AI AUTOMATION / DATA / RELIABILITY
I have been building AI/LLM-assisted software and automation since November 2022. I turn messy workflows into small, testable systems: AI agents, API integrations, data checks, automation recovery and practical internal tools you can run, inspect and hand off.
The general request desk stays broad; these focused pages show the exact delivery model and closest public proof for common work.
Agents, n8n/Make, APIs and operational workflows.
Open service ↗ FAILURE / RECOVERYRetries, timeouts, observability and duplicate prevention.
Open service ↗ JSON / CSVValidation, drift detection, cleanup and QA.
Open service ↗ PYTHONBrowser tasks, APIs, CLIs and small operational tools.
Open service ↗Direct clients can book the fixed scopes below. If we first met through Upwork or another marketplace, the project and payment stay on that marketplace.
For a workflow that works most of the time but fails in the ugly 10%.
For exports, API payloads or pipeline data that need repeatable validation.
For teams that need repeatable AI output instead of one-off prompts.
For one JSON API or webhook path that needs explicit validation before production or CI.
For webhook, API, AI-agent or automation steps where retrying can duplicate money or work.
Typical work includes webhook/API integrations, Google Workspace automation, AI-assisted classification or drafting, data validation, failure alerts, and small internal tools. I prefer one production path first, then expand after it survives real inputs.
Public repositories and demos are intentionally small enough to review.
Dependency-free JSON/JSONL drift detection with secret redaction, strict parsing, tests and CI failure gates.
Code ↗ WORKFLOW CONTRACTSExplicit API/webhook fixture contracts with secret-safe diagnostics, deterministic exit codes, regression tests and CI.
Code ↗ SIDE-EFFECT RELIABILITYSQLite receipts that prevent duplicate webhook, API and automation side effects with explicit execute/skip/conflict semantics.
Code ↗ AGENT RELIABILITYLong-running agent and external-effect patterns centered on state, receipts and duplicate prevention.
Code ↗ CONTENT SYSTEMSA validation-first static-page pipeline built from a structured brief, with safe URL boundaries and repeatable output.
Code ↗Define the input, expected result and failure condition before changing the system.
Run realistic inputs, capture actual outputs and keep failure behavior visible.
Leave concise setup, test and rollback notes so the work is maintainable after delivery.
A workflow diagram is nice, but a failing payload, screenshot, CSV or short written brief is enough to start scoping.