Where developers grow together

Alphabet - Intelligent Data Agent

What This Build Is About The Intelligent Data Dictionary Agent is an AI-powered, full-stack application designed to automate database documentation and facilitate natural language querying. Built with a Python/FastAPI backend and a React frontend, it connects seamlessly to any SQLAlchemy-supported database. Once connected, it automatically scans schemas, tables, and views to generate business-friendly summaries, key metrics, and plain-English schema explanations. The core of the application relies on local AI models (Llama 3 and Code Llama via Ollama), ensuring data privacy while powering an interactive chat interface. This chat features two modes: a "Summary Mode" for high-level metadata questions and a self-correcting "SQL Mode" that translates natural language into actionable database queries, executes them, and summarizes the results. Why I Built It Tackling the Intelligent Data Dictionary Agent track for the HackFest 2.0 hackathon by GDG Cloud New Delhi provided the perfect opportunity to build a high-impact, real-world tool. As a second-year CS undergrad actively exploring the data engineering space, I recognized that maintaining accurate database documentation and writing complex ad-hoc queries are massive bottlenecks for data teams. I wanted to build a solution that automates the tedious aspects of metadata management while giving users a smarter, faster way to interact with raw data without compromising on privacy. How It Can Be Useful to Others This tool provides immense value across different roles within an organization: Data Engineers & DBAs: Eliminates the manual, time-consuming process of writing and updating data dictionaries. The agent automatically keeps documentation in sync with the database structure. Non-Technical Stakeholders: Democratizes data access. Product managers, marketers, or executives can ask questions in plain English (e.g., "How many orders were placed last month?") without needing to know SQL or rely on an engineering team to pull the data. Security & Compliance Teams: Because the AI models (Llama 3 and Code Llama) run locally via Ollama, sensitive database schemas and query results never leave the local environment, ensuring strict data privacy and security.

CONTEXTO AI

"Enterprise data is dark."Every company relies on databases, but the documentation explaining them is usually outdated, non-existent, or trapped in technical jargon. When business teams need answers, they wait days for data engineers to translate cryptic column names like cust_ltv_q4_flg into actual insights. This documentation debt creates a massive bottleneck for decision-making. "Contexto AI is an autonomous data sense-making engine."We’ve built a platform that plugs directly into enterprise databases, extracts the raw schema, runs statistical health checks, and uses AI to automatically write business-ready documentation. We are turning static, confusing databases into an interactive, conversational asset where anyone can simply ask questions about their data. The Foundation (Backend & Extraction): We built a high-performance backend using Python and FastAPI for rapid, asynchronous processing. Using SQLAlchemy, we established secure, database-agnostic connectors that can ingest INFORMATION_SCHEMA metadata seamlessly from sources like PostgreSQL or Snowflake.The Intelligence (Google Gemini 1.5 Flash): Speed is critical for a conversational interface. We chose Gemini 1.5 Flash because its low-latency reasoning is perfect for instantly translating technical metadata into plain English and automatically tagging sensitive data (PII) on the fly.The Lab (Google AI Studio): To ensure accuracy, we utilized Google AI Studio to rapidly prototype, test, and refine our system prompts against complex database schemas before hardcoding the API calls into our backend.The Engine (RAG Pipeline): Standard LLMs hallucinate. To prevent this, we implemented a Retrieval-Augmented Generation (RAG) architecture. We embed the AI-generated documentation into a vector database. When a user asks a question, the AI only pulls from this highly-structured, verified metadata To prove this architecture works in a realistic, messy enterprise environment, we validated our extraction and AI enrichment tools against the GDG Cloud New Delhi Hackfest 2.0 dataset. By processing these internal recommendation documents and mock schemas, we successfully demonstrated the platform's ability to map complex data into searchable, business-ready insights. With Contexto AI, business users can finally ask, "Where is the Q4 revenue data?" and get an instant, accurate answer with a data trust score and suggested SQL. We are eliminating documentation debt and democratizing data access.

SmartOps

This project focuses on solving a common problem in real-world software development: important requirements are scattered across emails, chats, meetings, and documents, making them hard to track and easy to miss. The system automatically collects this communication data, removes irrelevant content, and identifies key business requirements, decisions, timelines, and risks. Instead of relying only on text generation, the project uses machine learning and rule-based logic to ensure the extracted information is accurate, explainable, and traceable to its original source. The final output is a structured Business Requirements Document that helps teams understand project goals clearly, detect conflicts early, and reduce manual effort in requirement analysis.