Build Without Limits - Lessons Learned Working With Data and AI Coding Assistants

© 2025, Amazon Web Services, Inc. or its affiliates. All
rights reserved. Amazon Confidential and Trademark. © 2025, Amazon Web Services, Inc. or its affiliates. All rights reserved. Amazon Confidential and Trademark. Lessons Learned Working With Data and AI Coding Assistants Ricardo Sueiras (vi/vim) B U I L D W I T H O U T L I M I T S Developer Advocate, AWS

rights reserved. Amazon Confidential and Trademark. A picture of Sir Isaac Newton standing on the shoulders of giants. pixel art (8 bit) with a blue colour theme.

rights reserved. Amazon Confidential and Trademark. Developer Velocity Assembly Language Machine Code Levels of abstraction Punch Cards Compiled Languages Interpreted Languages

rights reserved. Amazon Confidential and Trademark. Autocompletion Tools Frameworks (code generation) Domain Specific Languages IDE Developer Velocity Sophistication of tooling Version Control Profilers Compilers Linters

rights reserved. Amazon Confidential and Trademark. Each one helps us to do a better job and to write better code

rights reserved. Amazon Confidential and Trademark. Search Engines Community knowledge bases (Stack Overflow) Pre-internet Supporting infrastructure and systems Developer Velocity

rights reserved. Amazon Confidential and Trademark. Search Engines Community knowledge bases Pre-internet Supporting infrastructure and systems Developer Velocity Generative AI

rights reserved. Amazon Confidential and Trademark. You are (so) wrong!

rights reserved. Amazon Confidential and Trademark. You are absolutely right!

rights reserved. Amazon Confidential and Trademark. AI Coding Assistants provide a GUI on top of large language models, that optimize use for developers across the software development life cycle (SDLC)

rights reserved. Amazon Confidential and Trademark. Binary Code 00100110 10010011 … 11000011 00111100 Assembly Language LDA #20 ADC #50 STA RESULT Assembler (Tool) High-Level Language int A, B, RESULT; A = 20; B = 50; RESULT = A + B; Compiler (Tool) Ai-Powered Coding Assistant “Start with integers A and B, add them and store as RESULT.” AI-Assistant / LLM (Tools)

rights reserved. Amazon Confidential and Trademark.

rights reserved. Amazon Confidential and Trademark. Emerging patterns Exploration Learning and Research Developer flow

rights reserved. Amazon Confidential and Trademark. Exploration • Reduce friction • Do the unexpected • Rapid prototyping

rights reserved. Amazon Confidential and Trademark. Learning Research • Learn new concepts • Critic and reviewer • How does this work?

rights reserved. Amazon Confidential and Trademark. Developer flow • Automation • Reduce context switching • Debugging and stack trace analysis • Situational awareness • Refactor and migration

rights reserved. Amazon Confidential and Trademark. Can you use AI Coding Assistants when working with data?

© 2025, Amazon Web Services, Inc. or its aﬃliates. All
rights reserved. Amazon Conﬁdential and Trademark. AGENTS.md --- description: Comprehensive best practices and coding standards for utilizing Pydantic effectively in Python projects --- - **Model Definition:** - Use `BaseModel` to define data schemas with type annotations for clarity and automatic validation. - Prefer simple models that encapsulate a single concept to maintain readability and manageability. - Use nested models for complex data structures while ensuring each model has clear validation rules. - Always define a `Config` class within your model to control model behavior. - **Validation and Error Handling:** - Implement built-in and custom validators to enforce data integrity. - Utilize `@field_validator` for field-specific rules and `@root_validator` for cross-field validation. - Ensure that validation errors are user-friendly and logged for debugging purposes. - Custom error messages should be informative and guide the user on how to correct the data. - Use `ValidationError` to catch and handle validation errors.

rights reserved. Amazon Conﬁdential and Trademark. sql-standards.md # SQL Development Guidelines This document provides a comprehensive set of guidelines for SQL development, covering aspects from naming conventions and table design to query optimization, security, and testing, aimed at promoting best practices in database management and application development. ## Implementing These Principles These guidelines are built upon the following core principles: - Principle 1: Clarity and Maintainability: Ensuring SQL code and database structures are easy to understand, modify, and maintain. - Principle 2: Performance and Scalability: Designing databases and queries for optimal performance and the ability to handle growing data and user loads. - Principle 3: Data Integrity and Security: Protecting data accuracy, consistency, and safeguarding against unauthorized access or loss. - Principle 4: Consistency: Applying uniform standards across the database schema and SQL code. ## Table of contents - Rule 1: Naming Conventions - Rule 2: Table Design - Rule 3: Query Writing - Rule 4: Indexing Strategy - Rule 5: Security Guidelines - Rule 6: Performance Optimization Tips - Rule 7: Transaction Guidelines

rights reserved. Amazon Conﬁdential and Trademark. ├── agents │ ├── database-planner │ │ └── Agent.md │ ├── database-architect │ │ └── Agent.md │ ├── sql-reviwer │ │ └── Agent.md │ └── sql-developer │ └── Agent.md Your Role: You are an experienced database architect. Before you start any task, please do the planning and write your steps in a “data_model_plan.md” file with checkboxes against each step in the plan. If any step needs my clarification, please add it to the step to interact with me and get my confirmation. Do not make critical decisions on your own. Once you produce the plan, ask for my review and approval. After my approval, you can go ahead to execute the same plan one step at a time. Once you finish each step, mark the checkboxes as done in the plan. database-architect

rights reserved. Amazon Conﬁdential and Trademark. ├── agents │ ├── database-planner │ │ └── Agent.md │ ├── database-architect │ │ └── Agent.md │ ├── sql-reviwer │ │ └── Agent.md │ └── sql-developer │ └── Agent.md Your Role: You are an experienced SQL developer. You have knowledge of all SQL dialects. Use these guidelines and standards when writing SQL code SQL Formatting Rules: - Use UPPERCASE for SQL keywords (SELECT, FROM, WHERE, JOIN, etc.) - Use 4 spaces for indentation - Each SELECT column on a new line with comma at the beginning - JOIN clauses indented and aligned - WHERE conditions on separate lines with logical operators at the beginning - Proper spacing around operators (=, >=, <=, etc.) - CASE/WHEN clauses properly indented with each condition on new line - Subqueries properly indented and parentheses aligned - Comments formatted consistently sql-developer

rights reserved. Amazon Conﬁdential and Trademark. database-architect postgresql Your Role: You are an experienced database architect. Before you start any task, please do the planning and write your steps in a “data_architecture_plan.md” file with checkboxes against each step in the plan. If any step needs my clarification, please add it to the step to interact with me and get my confirmation. Do not make critical decisions on your own. Once you produce the plan, ask for my review and approval. After my approval, you can go ahead to execute the same plan one step at a time. Once you finish each step, mark the checkboxes as done in the plan. # PostgresSQL rules ## General - Use lowercase for SQL reserved words to maintain consistency and readability. - Employ consistent, descriptive identifiers for tables, columns, and other database objects. - Use white space and indentation to enhance the readability of your code. - Store dates in ISO 8601 format (`yyyy-mm- ddThh:mm:ss.sssss`). - Include comments for complex logic, using '/* ... */' for block comments and '--' for line comments. ## Naming Conventions - Avoid SQL reserved words and ensure names are unique and under 63 characters. - Use snake_case for tables and columns. - Prefer plurals for table names - Prefer singular names for columns.

rights reserved. Amazon Conﬁdential and Trademark. Use AI coding assistant to create initial steering documents

rights reserved. Amazon Conﬁdential and Trademark. Setup MCP to connect to your databases

rights reserved. Amazon Conﬁdential and Trademark. https://playbooks.com/mcp https://context7.com

rights reserved. Amazon Conﬁdential and Trademark. https://github.com/awslabs/mcp MCP Servers for working across all AWS data services

rights reserved. Amazon Conﬁdential and Trademark. Treat MCP servers like you would any 3rd library : potentially harmful and hostile

rights reserved. Amazon Conﬁdential and Trademark. # Secure SQL Usage These rules apply to all code that interacts with SQL databases, regardless of language or framework, including generated code. Generated code must not violate these rules. If a rule is violated, a comment must be added explaining the issue and suggesting a correction. ## 1. Always Use Parameterized Queries and Prepared Statements - **Rule:** Never construct SQL queries by concatenating or interpolating user input directly into the query string. Always use parameterized queries or prepared statements provided by your database library. Prepared statements ensure that user input is treated as data, not executable code, and are the most effective way to prevent SQL injection. - **Example (unsafe, Python):** ```python cursor.execute(f"SELECT * FROM users WHERE username = '{username}'") ``` - **Example (safe, Python):** ```python cursor.execute("SELECT * FROM users WHERE username = ?", (username,)) ``` - **Example (unsafe, PHP):** ```php $query = "SELECT * FROM users WHERE username = '" . $_GET['username'] . "'"; $result = mysqli_query($conn, $query); ``` - **Example (safe, PHP - using prepared statements):** ```php $stmt = $conn->prepare("SELECT * FROM users WHERE username = ?"); $stmt->bind_param("s", $_GET['username']); $stmt->execute(); $result = $stmt->get_result(); ```

rights reserved. Amazon Conﬁdential and Trademark. ## 2. Validate and Sanitize All Input - **Rule:** All external input used in queries must be validated for type, length, and format. Apply allow-lists where possible. ## 3. Avoid Dynamic SQL Unless Absolutely Necessary - **Rule:** Do not use dynamic SQL (e.g., building table or column names from user input) unless there is no alternative. If unavoidable, strictly validate and allow-list all dynamic parts. ## 4. Use Least Privilege Database Accounts - **Rule:** Connect to the database using accounts with the minimum privileges required for the task. Never use admin or superuser accounts for application queries. ## 5. Do Not Expose Database Errors to Users - **Rule:** Never return raw database error messages to end users. Log errors securely and show generic error messages instead. ## 6. Avoid Storing Sensitive Data in Plaintext - **Rule:** Do not store passwords, tokens, or sensitive data in plaintext. Use strong, salted hashing for passwords and encryption for sensitive fields. ## 7. Close Database Connections Properly - **Rule:** Always close database connections and cursors to avoid resource leaks and potential security issues.

rights reserved. Amazon Confidential and Trademark. © 2025, Amazon Web Services, Inc. or its affiliates. All rights reserved. Amazon Confidential and Trademark. • Be frugal with MCP servers • Edit your rules files ruthlessly • Context is king! • Tailor context for specific tasks • Create (multiple) custom agents (personas) for specific data activities • Use existing SOPs, documentation, or knowledgebases to help seed these • Define controls What works for me…

rights reserved. Amazon Conﬁdential and Trademark. 35 Provide a chat interface to your data

rights reserved. Amazon Conﬁdential and Trademark. Provide a chat interface to your data

rights reserved. Amazon Conﬁdential and Trademark. Simplify working across different data formats

rights reserved. Amazon Conﬁdential and Trademark. Simplify generating schemas from sample data

rights reserved. Amazon Conﬁdential and Trademark. Debugging issues with data formats

rights reserved. Amazon Conﬁdential and Trademark. a new data format that cuts LLM token usage by 30–60% when feeding in structured data. Token-Oriented Object Notation (TOON)

rights reserved. Amazon Conﬁdential and Trademark. 41 This space is changing so rapidly, with new updates happening on an almost weekly basis – make sure you keep up to date with the news!

rights reserved. Amazon Conﬁdential and Trademark. Transform initial data set (500 images) into separate image and csv file to build a simple game

rights reserved. Amazon Conﬁdential and Trademark. © 2025, Amazon Web Services, Inc. or its affiliates. All rights reserved. Amazon Confidential and Trademark. • Provide examples and additional context when working with new data formats • Validate/test schemas • Balance cost optimisation, efficiency, and readability when deciding data formats • Use custom personas to set data format preferences • Explore MCP Servers Some solid pointers…

rights reserved. Amazon Conﬁdential and Trademark. ├── agents │ ├── database-planner │ │ └── Agent.md │ ├── test-data │ │ └── Agent.md │ ├── sql-reviwer │ │ └── Agent.md │ └── sql-developer │ └── Agent.md You will be asked to create various types of synthetic data to use for testing systems. You may be asked for a variety of formats, but more than likely you will be asked for XML, JSON, CSV, JSONL, or plain text. To generate good data, introduce yourself as "Orac," and that you are tailored to creating synthetic data. To do so you will guide the user through a simple qna (defined within the <qna/> XML tags). <qna> 1. "Hello! Do you have data to use as an example? If so please upload or paste that data example into the chat and send it to me." 1.a. If they do have data use it as an example for your work and store it in your <agent_scratchpad/> XML tags for your use. 1.b. If no example try and ask them for the target data format, example: txt, xml, json, etc. 1.c If no example data move to question 2. 2. "What domain or "theme" should the data be modeled for? Examples: Medical, Geographic, Technology" 3. "Do you have a desired schema or data format the synthetic data must adhere to? If so please upload/paste/send the information to me." 3.a. If they have an example then store it in your <agent_scratchpad/> XML tags. 3.b If they don't have an example ask them if they want you to try your best to create a normalized form of it. 4. "Do you have specific fields that you need in the generated synthetic data I generate?" 4.a. If they have specific fields, store them in your <agent_scratchpad/> XML tags. 5. "Thank you, are you ready for me to proceed?" </qna> test-data-generator

rights reserved. Amazon Conﬁdential and Trademark.

rights reserved. Amazon Conﬁdential and Trademark. Simplify working with test data generators like faker

rights reserved. Amazon Conﬁdential and Trademark. https://github.com/brightdata/brightdata-mcp

rights reserved. Amazon Confidential and Trademark. © 2025, Amazon Web Services, Inc. or its affiliates. All rights reserved. Amazon Confidential and Trademark. • Create custom agents • Provide examples of the sample data you want to generate • Generate good and bad data • Use context to help shape • Explore MCP servers available (or create your own) What works well

rights reserved. Amazon Conﬁdential and Trademark. AI coding assistants are great natural language interfaces into your data

rights reserved. Amazon Conﬁdential and Trademark.

rights reserved. Amazon Conﬁdential and Trademark. Explaining Writing Querying Debugging • Understand SQL code you have been given • Onboarding new SQL developers • Help with SQL reviews • Writing documentation and READMEs • Simplify writing queries • Simplify refactoring SQL • Provide a text interface to access your data • Accelerate building data applications • Fast iteration • Fix and debug development issues • Identify operational problems

rights reserved. Amazon Confidential and Trademark. Explaining

rights reserved. Amazon Confidential and Trademark. 56 Explaining

rights reserved. Amazon Confidential and Trademark. select u.user_id,u.username,u.email,p.product_name,o.order_date,o.total_amount from users u join orders o on u.user_id=o.user_id join order_items oi on o.order_id=oi.order_id join products p on oi.product_id=p.product_id where o.order_date>=date_sub(now(),interval 30 day) and o.status='completed' order by o.order_date desc,u.username asc; Explaining

rights reserved. Amazon Confidential and Trademark. ├── agents │ ├── database-planner │ │ └── Agent.md │ ├── test-data │ │ └── Agent.md │ ├── sql-reviwer │ │ └── Agent.md │ └── sql-developer │ └── Agent.md SQL Formatting Rules: - Use UPPERCASE for SQL keywords (SELECT, FROM, WHERE, JOIN, etc.) - Use 4 spaces for indentation - Each SELECT column on a new line with comma at the beginning - JOIN clauses indented and aligned - WHERE conditions on separate lines with logical operators at the beginning - Proper spacing around operators (=, >=, <=, etc.) - CASE/WHEN clauses properly indented with each condition on new line - Subqueries properly indented and parentheses aligned - Comments formatted consistently sql-reviewer Explaining

rights reserved. Amazon Confidential and Trademark. SELECT u.user_id, u.username, u.email, p.product_name, o.order_date, o.total_amount FROM users u JOIN orders o ON u.user_id = o.user_id JOIN order_items oi ON o.order_id = oi.order_id JOIN products p ON oi.product_id = p.product_id WHERE o.order_date >= DATE_SUB(NOW(), INTERVAL 30 DAY) AND o.status = 'completed' ORDER BY o.order_date DESC, u.username ASC; Explaining

rights reserved. Amazon Confidential and Trademark. Writing https://pub.towardsai.net/the-sql-generation-showdown-testing-top-ai-models-58e547781724

rights reserved. Amazon Confidential and Trademark. Writing Amazon Redshift query editor now has Amazon Q Developer integrated to help you write queries

rights reserved. Amazon Confidential and Trademark. Debugging AI Coding Assistants are not going to make all the bugs vanish into thin air - but it does change how you handle them

rights reserved. Amazon Confidential and Trademark. Debugging

rights reserved. Amazon Confidential and Trademark. Querying Developing your own text to SQL interface has never been easier

rights reserved. Amazon Confidential and Trademark. {"eventType": "ReferralPath", "repo_name": "094459-amazon-q-git-demo", "path": "/094459/094459-amazon-q-git-demo", "title": "GitHub - 094459/094459-amazon-q-git-demo: A simple Java app that shows how Am...", "count": 3, "uniques": 2} {"eventType": "View", "repo_name": "094459-amazon-q-git-demo", "count": 6, "uniques": 3} {"eventType": "Clone", "repo_name": "094459-amazon-q-git-demo", "count": 5, "uniques": 4} https://dev.to/aws/how-i-used-amazon-q-developer-to-move-faster-with-data-3e7i Querying

rights reserved. Amazon Confidential and Trademark. https://bird-bench.github.io/ A SQL equivalent of SWE Bench

rights reserved. Amazon Confidential and Trademark. © 2025, Amazon Web Services, Inc. or its affiliates. All rights reserved. Amazon Confidential and Trademark. • Use prompting and context engineering practices • Write queries to build context • Break down complex queries into a series of tasks • Start with intent and then provide context • Iterate and refine • Provide the right context and domain specific knowledge • Include database schema and query plan for the best results Query these..

rights reserved. Amazon Confidential and Trademark. Generate data from ERD

rights reserved. Amazon Confidential and Trademark. 71

rights reserved. Amazon Confidential and Trademark. Generate ERD from data https://www.kaggle.com/datasets/jsphyg/star-wars

rights reserved. Amazon Confidential and Trademark. MCP Servers provide diagram generation capabilities across several different diagram types

rights reserved. Amazon Confidential and Trademark. © 2025, Amazon Web Services, Inc. or its affiliates. All rights reserved. Amazon Confidential and Trademark. • ALWAYS validate and check generation • Use Mermaid format and markdown for portability • Use ERD diagrams as context • Explore available MCP Servers for generating and understanding diagrams Things to think about

rights reserved. Amazon Confidential and Trademark. Migrate Apache Airflow DAGs to the latest version of Apache Airflow

rights reserved. Amazon Confidential and Trademark. Migrating Redis to Valkey and upgrading to the latest GLIDE client code

rights reserved. Amazon Confidential and Trademark. /transform migrates your Java applications AND your SQL statements

rights reserved. Amazon Confidential and Trademark. oeddit, an open source PHP reddit clone was refactored to Python, including migrating from MySQL to PostgreSQL

rights reserved. Amazon Confidential and Trademark. https://github.com/j4c0bs/mcp-server-sql-analyzer A Model Context Protocol (MCP) server that provides SQL analysis, linting, and dialect conversion capabilities using SQLGlot.

rights reserved. Amazon Confidential and Trademark. © 2025, Amazon Web Services, Inc. or its aﬃliates. All rights reserved. Amazon Conﬁdential and Trademark. • Provide enough example SQL within your steering / context files • Implement a testing/validation approach to help with review • Some LLMs are better than others when it comes to SQL – constantly review What I have found that works

rights reserved. Amazon Confidential and Trademark. Add your data model and then conﬁgure context

rights reserved. Amazon Confidential and Trademark. Spec driven development uses data models to guide code generation

rights reserved. Amazon Confidential and Trademark. https://github.com/centralmind/gateway MCP Server that helps you automatically build APIs from your data

rights reserved. Amazon Confidential and Trademark. © 2025, Amazon Web Services, Inc. or its affiliates. All rights reserved. Amazon Confidential and Trademark. • Use data models to build strong grounding for your AI coding assistant • Update code from data model (and vice versa) • spec driven development leverages data models What works well

rights reserved. Amazon Confidential and Trademark. Use observability data from your database to guide suggestions and optimisation recommendations

rights reserved. Amazon Confidential and Trademark. ├── agents │ ├── database-planner │ │ └── Agent.md │ ├── test-data │ │ └── Agent.md │ ├── sql-optimiser │ │ └── Agent.md │ └── sql-developer │ └── Agent.md sql-optimiser # SQL and Data Optimization Rules You are a SQL Optimising agent. You are an expert in SQL, and understand all dialects. Your role is to reviewing SQL code against four key criteria: query design, index design, performance, and security. ## Review Process ### 1. Query Design Review #### Evaluation Criteria - **Readability & Maintainability** - Table aliases: 2-4 characters, meaningful (u for users, o for orders) - Column selection: Explicit columns only, no SELECT * - Indentation: 2-4 spaces, aligned JOIN/WHERE clauses - Line length: Max 120 characters per line - **Correctness & Logic** - JOIN conditions: Always include ON clause, avoid comma joins - WHERE clauses: Use proper operators (= vs LIKE vs IN) - NULL handling: Explicit IS NULL/IS NOT NULL checks - Aggregates: GROUP BY includes all non-aggregate SELECT columns - Data types: Consistent comparison types (avoid implicit conversions) - **Query Structure Efficiency** - CTEs for readability when query >3 levels deep - Window functions instead of self-joins for ranking - EXISTS instead of IN for subqueries with potential NULLs - UNION ALL instead of UNION when duplicates acceptable

rights reserved. Amazon Confidential and Trademark. https://github.com/crystaldba/postgres-mcp MCP Servers can be conﬁgured to connect directly to your databases and provide you with optimisation guidance and suggestions

rights reserved. Amazon Confidential and Trademark. © 2025, Amazon Web Services, Inc. or its affiliates. All rights reserved. Amazon Confidential and Trademark. • Create custom agents that are experts in optimising your SQL • Provide examples of what good looks like • Provide operational observability data as additional context • Explore MCP Servers Dive deeper

rights reserved. Amazon Confidential and Trademark. What I have learned…. …just scratched the surface of what you can do

rights reserved. Amazon Confidential and Trademark. Make sure you know how your AI coding assistant manages “data” that you are working with Privacy Review Stay up to date Half life of innovation is so fast, make sure you stay up to date and make sure you stay in the loop! AI Coding Assistants hallucinate, go off rails, and are not detail orientated – so ALWAYS review and check

rights reserved. Amazon Conﬁdential and Trademark. The next generation AI coding assistant, that works across IDE and terminal https://kiro.dev/

rights reserved. Amazon Conﬁdential and Trademark. https://s12d.com/builder-id AWS Builder Centre Lots of great Kiro content to get you started

rights reserved. Amazon Conﬁdential and Trademark. Thank you! © 2025, Amazon Web Services, Inc. or its affiliates. All rights reserved. Amazon Confidential and Trademark. Ricardo Sueiras | Developer Advocate [email protected] https://pulse.aws/survey/FMMUY4TB

Build Without Limits - Lessons Learned Working ...

Build Without Limits - Lessons Learned Working With Data and AI Coding Assistants

More Decks by Ricardo Sueiras

Featured

Transcript