Description
Summary We’re building a focused proof of concept for Pipeline Sentinel — an agentic AI copilot for enterprise data teams that watches warehouse and pipeline health, explains failures in plain language, and proposes (or drafts) remediation steps without becoming a full product yet. The idea Modern data orgs drown in fragmented signals: dbt test failures, Airflow/DAG alerts, Snowflake query errors, and Slack noise. Pipeline Sentinel’s POC should prove one end-to-end loop: ingest a small set of real or realistic failure events, use a multi-agent orchestration layer to diagnose root cause across lineage + logs + recent schema/query changes, then surface a structured “incident brief” in a simple web UI (what broke, why, blast radius, suggested fix, confidence). Success looks like a credible demo on a scripted incident pack — not a polished SaaS MVP. You’ll own a thin vertical slice across the stack: event/log ingestion and light transformation into an analytics/serving layer, agent orchestration against enterprise model APIs, retrieval over runbooks/lineage snippets, and a clean operator UI to trigger investigations and review outputs. Keep scope tight: one domain (e.g. dbt + warehouse freshness/quality failures), a handful of agents (triage, lineage/context, remediation draft), and a demo path we can walk in a Loom. This is a POC only — architecture should be production-minded and enterprise-credible, but we are not asking for auth hardening, multi-tenant SaaS, full observability platform, or MVP feature breadth. Strong delivery may lead to a paid next phase (deeper connectors, evaluation harness, and hardening). What “good” looks like A working demo where an operator picks (or auto-detects) an incident, agents run with visible traces/tool calls, and the UI returns a decision-ready brief with citations to the evidence used. Code should be clean, repo-ready, and easy for us to extend. Must-have application materials Apply only if you have directly similar experience shipping agentic systems on enterprise data stacks (not toy chatbots). In your proposal, include: - 2–3 past projects closest to this (agent + warehouse/pipeline + UI), with your exact role - Links to a GitHub portfolio (or private repo access on request) showing production-quality structure - A short Loom (≤5 min) walking a similar build: architecture, agent/tool design, and a live or recorded demo Proposals without Loom + GitHub + relevant case detail will be skipped.