Rajan Mishra
Architecting production RAG pipelines and autonomous multi-a
- 1
- Followers
"Can I upgrade our payment service from v2 to v4 tonight?"
Engineers propose production changes like this based on partial mental models. The facts that invalidate the plan are already documented ā but scattered across release notes, migration guides, and stale architecture docs.
Standard AI tools fail catastrophically here.
A typical RAG chatbot uses semantic similarity search over raw text chunks. But similarity search has NO concept of:
ā Relational dependency chains ("v4 requires SDK v3, but production runs SDK v2")
ā Authority hierarchy ("Release Notes supersede stale architecture docs")
ā Policy exceptions ("Emergency hotfixes bypass 24h staging soak rules")
To solve this, I built SHIPCHECK:
š "Before you ship a change, let the agent try to break your plan."
Instead of treating documentation like an unorganized soup of text embeddings, ShipCheck models production facts as a structured knowledge graph queried with deterministic GROQ graph traversal:
1ļøā£ Dependency Traversal: Maps microservice relationships and version constraints deterministically.
2ļøā£ Authority Weighting: Prioritizes active incident postmortems and verified release notes over stale wiki pages.
3ļøā£ Adversarial Verification: The agent actively tries to invalidate your deployment proposal before it touches staging or prod.
Structure beats plain retrieval every single time.
š» Check out the open-source repo: https://github.com/aiwithrajan/shipcheck
āāāāāāāāāāāāāāāāāāāāāāāāāā
I'm an AI Systems & Agent Architect and Verified GitHub Sponsor.
If your engineering team or startup wants to move beyond naive RAG chatbots to build deterministic verification agents, knowledge graphs, or custom LangGraph workflows:
Drop me a DM or connect on GitHub! Let's build something unbreakable. š ļø
#DevOps #PlatformEngineering #AI #SoftwareEngineering #BuildInPublic #OpenSource #Python
2
0
74
Zero-Hallucination Document Intelligence & RAG Agent
0
19