Cortex Code Security: Protecting IP with Governance

If you are using Snowflake and have not tested Cortex Code yet, Nick Akincilar explains in his post why CoCo is the superior choice. A must read if you still haven't taken advantage of this gamechanger feature. ❄️

"It works perfectly on my machine." Then your teammate clones the repo, runs 'pnpm start:dev', and immediately hits a wall: ❌ PrismaClientInitializationError: Can't reach database server ❌ The table 'public.User' does not exist ❌ No default admin account to log in with I used to run into this constantly. Every time I handed a NestJS project to someone else (or revisited it myself months later), the setup was a nightmare. To fix this, I completely standardized my backend setup flow using Docker, Prisma, and an Idempotent Seed Script. Now, anyone can clone my repos and have a fully functioning environment in just a few commands. Here is the exact architecture I use to eliminate setup friction: 1. Isolate the Database with Docker Compose Stop installing PostgreSQL directly on your machine. A simple docker-compose file pulling the Postgres 16 image ensures every developer is running the exact same database version. Use volumes so data survives container restarts. 2. Automate Schema Syncing Instead of writing manual SQL, rely on Prisma. For solo/early projects: "npx prisma db push" For team/production apps: "npx prisma migrate dev" This ensures the database structure always matches your TypeScript models. 3. Write "Idempotent" Seed Scripts This is the step most people skip. Write a seed script that automatically generates a default Admin account and essential app data. Crucial detail: Always use Prisma’s Upsert method instead of Create. This means "if the user exists, do nothing; if not, create it." This makes it safe to run the seed script multiple times without duplicating data. 4. Fix the Build Order Trap A common CI/CD failure happens when NestJS compiles before Prisma generates its client types. The fix? Update your package.json build script to: "build": "prisma generate && nest build" Having a clean, repeatable setup flow is the difference between a frustrating "works on my machine" project and a professional, production-ready codebase. I wrote a comprehensive, step-by-step breakdown of this exact setup, including code snippets for the Dockerfile and the seed script. Check the first comment for the link to the full guide! 👇 How do you handle local database setups for your teams? Do you prefer Docker or cloud-hosted dev databases? #NestJS #Prisma #Docker #BackendEngineering #DevOps #SoftwareArchitecture

I'm using my first post here to share my experience with Snowflake Cortex Code. Cortex Code is an AI coding assistant that runs directly in Snowsight. I was curious whether it's actually useful when you try to build something real, with the constraint of staying entirely within the platform. The use case: an AI agent that can learn in a controlled manner. The scenario: a user asks the agent something, the agent doesn't have an answer. The user says "Remember this: X." But that shouldn't flow into the knowledge base unchecked. So there needs to be a review process in between. The Architecture User → Agent → Feedback Queue → Review (Streamlit) → Knowledge Base → Cortex Search → Agent Is this the optimal architecture? Debatable, but this is about technical feasibility, and sometimes that means being pragmatic. Technical Details The agent has two tools: knowledge_search (Cortex Search over the knowledge base) and submit_feedback (a SQL procedure that writes feedback to a queue). When a user provides feedback, submit_feedback stores not just the content but also the context, the original question and the agent's response. Plus a feedback_type (new_info or correction) and a category. Everything lands in FEEDBACK_QUEUE with status = 'pending'. In the Streamlit review dashboard, the reviewer sees the feedback with full context. Duplicate detection runs via Cortex Embed using VECTOR_COSINE_SIMILARITY against existing entries. The threshold is configurable. The reviewer has three options: Approve adds the entry to KNOWLEDGE_BASE with source_feedback_id for traceability. Reject sets the status to rejected. Replace soft deletes the old entry (is_active = FALSE) and adds the new one in a single step. Cortex Search runs with TARGET_LAG = 1 minute and filters on is_active = TRUE. After approval or replacement, the agent picks up the new knowledge within a minute. The result: two tables, one Cortex Search Service, five SQL procedures, one agent, one Streamlit app. All within a single platform, as intended. On Cortex Code itself: Did I write most of the code myself, or did I just vibe code my way through? We may never know. What I can say is that it was genuinely helpful, especially with platform specific syntax like Cortex Search DDL or agent configuration. Open Questions This PoC keeps things simple. The natural next questions would be: How do you prevent the queue from being flooded (rate limiting)? What happens when knowledge originates from external systems and drifts from the source of truth? Should there be quality signals like confidence scores before feedback reaches the review queue? And is soft delete enough for rollback, or do you need full versioning? Curious to hear how others approach these challenges.

🚀 Built a Baseline Enterprise RAG Assistant (Open-Source Stack) Over the past few weeks, I built a baseline Retrieval-Augmented Generation (RAG) system using a completely open-source stack. This is an MVP — and there’s a lot more I want to improve. 🏗 Current Stack • Java + Spring Boot • PostgreSQL + pgvector (HNSW indexing) • Ollama • Mistral (LLM) • nomic-embed-text (768-dim embeddings) 🔍 What the MVP Does • Ingests structured documents • Splits content into chunks • Generates embeddings • Stores vectors in PostgreSQL • Uses cosine similarity (<->) for semantic retrieval • Generates grounded answers using retrieved context Everything runs locally. No paid APIs. ⸻ 🧠 What I’m Planning Next I want to evolve this into a self-improving enterprise knowledge assistant: 🔹 Semantic chunking instead of fixed chunk sizes 🔹 PDF/eBook ingestion pipeline 🔹 Employee query feedback loop 🔹 Admin-approved knowledge updates 🔹 Role-based knowledge filtering 🔹 Confidence scoring & hallucination control Goal: A system that continuously improves as employees contribute corrections and internal documentation. Not model retraining — but structured knowledge evolution. ⸻ This is just the baseline architecture. There are several areas to improve — retrieval precision, chunking strategy, prompt design, evaluation metrics. I’d genuinely appreciate feedback from engineers working on RAG / enterprise AI systems and we will evolve brick by brick. 🔗 GitHub: https://lnkd.in/g7gq--gT #Java #SpringBoot #GenAI #RAG #OpenSource #LLM #BackendEngineering #PostgreSQL EY IBS Software Bosch Global Software Technologies KPMG US QBurst Experian Experion Technologies KeyValue athenahealth HashedIn by Deloitte UST HealthProof UST Thomson Reuters #GenAI #ArtificialIntelligence #MachineLearning #LLM #RAG #JavaDeveloper #SpringBoot #BackendEngineering #SoftwareEngineering #TechCareers #Hiring #OpenToWork #AIEngineering #Innovation

Excited to Share My Data Structures Repository on GitHub! As part of my continuous journey in software development and problem solving, I created a GitHub repository where I implement and practice fundamental Data Structures. Data Structures are the backbone of efficient algorithms and scalable systems. They help organize and store data efficiently so that operations like searching, inserting, and deleting can be performed effectively. (Gist) In this repository, I focus on implementing core concepts such as: Arrays Linked Lists Stack Queue Trees Algorithms related to these structures The goal of this project: Strengthen my problem-solving skills Understand how data structures work internally Build a strong foundation for software engineering and technical interviews I believe mastering data structures is essential for any developer who wants to write efficient and scalable code. Check out the repository here: https://lnkd.in/dAf3tRnT If you find it helpful, feel free to star the repository, share feedback, or contribute. #DataStructures #Algorithms #GitHub #Programming #SoftwareEngineering #ProblemSolving #Cplusplus #OpenSource #LearningJourney

I look back very fondly on my time as a backend developer. But looking back also makes it pretty obvious what things could've been done better. When Databricks Lakebase and its instant database branching were announced at our Data + AI Summit last year, I immediately thought: this would have saved me so much pain. In my previous role, the most frustrating parts of shipping weren’t the features themselves, it was the migration headaches around them. Coordinating schema changes with deployments, resolving merge conflicts, cherry-picking migration files, and rolling back when something did go wrong. In this writeup, I document my thoughts on why Lakebase branching is such a game changer for increasing Postgres database migration assurance, and walk through a practical example of how it can be integrated into a local Docker-based development workflow. https://lnkd.in/gY6JQ_Bs

Instant startup and database branching with Databricks Lakebase is an absolute game changer for devs that use Postgres for their backend. Being able to test database changes instantly against a complete branch of the production database and then shut down the branch the second the change is tested, is an amazing feature for checking your DB changes simply, cheaply and instantly! As a Developer I always wished I had features like this available for my CI/CD process when I was deploying large production apps. Give it a try today!

Automating Oracle Data Transforms just got a lot easier. In this new blog, with my colleague Ricardo Morais we show how to build a CI/CD pipeline around Oracle Data Transforms using REST APIs. If you are modernizing data pipelines on Autonomous Database or looking to bring DevOps practices to your ELT processes, this walkthrough is a great place to start. #oracle #oracledatatransforms #etl #python

What is CoCo? Why does it sound fancy? CoCo is Cortex Code in Snowflake. Either you can use UI or CLI to use it. But what is CoCo? Wait! Let me deep dive a bit. Cortex Code is an AI-driven intelligent agent integrated into the Snowflake platform. [INTELLIGENCE] It understands Snowflake. Think of it like a built-in expert on Snowflake. It has up-to-the-minute knowledge of your specific data catalog: databases, schemas, semantic models, and more. [RELEVANCE] It is context aware. The agent knows where you are in the UI, what data you're working with, and retains your conversational history as you navigate. [INTEGRATION] Automates complex tasks. Integrates natively with your stack: dbt, Git, SQL, Python, and Streamlit. No changes to how you work. [GOVERNANCE] Secure by design. Cortex Code uses Snowflake RBAC, ensuring it only sees and acts on the data and objects you have access to. I personally use it as a code reviewer, and it knows how to optimize where I may have missed something. It also helps me a lot with admin tasks. It has: CUSTOMIZE [AGENTS.md] - Provide context and instructions to help Cortex Code work on your project. - Bring AGENTS.md files from other agents and IDEs. CONNECT [MCP Servers] - Connect MCP servers to Cortex Code to work with other systems. - Connect JIRA, GitHub, Slack, and more. EXTEND [AGENT SKILLS] - Out-of-the-box skills for working with Snowflake. - Build skills for your team and share them. - Extend with publicly available skills. #snowflakesquad #snowflakecommunity #dataengineers #analytics #agenticAI Gregory Goomishian Elsa Mayer Amilee Alesna

🚀 Production-Style ETL Pipeline with Docker Compose I’m excited to share a small but production-focused ETL pipeline I built using Python, PostgreSQL, Docker, and Docker Compose to practice real-world data engineering patterns. Instead of just running a local script, I focused on building it the right way: ✅ Multi-stage Docker builds ✅ PostgreSQL health checks ✅ depends_on with service_healthy ✅ Environment-based configuration using .env ✅ Persistent storage with Docker volumes ✅ Running containers as a non-root user The ETL container connects to PostgreSQL, creates a table if it doesn’t exist, and inserts a sample record — fully containerized and reproducible with a single command. This project helped me move from simple script-level ETL to containerized, production-ready pipeline design.production-ready pipeline thinking. 🔗 GitHub Repository: https://lnkd.in/eieX_PqP Always building stronger data engineering systems. #DataEngineering #Docker #ETL #PostgreSQL #DevOps #Python #Backend #Cloud — Sri Krishna Sai Kota