Sanchit Agrawal
AI Solutions Engineer | AI Automation, AI Apps, Voice & RAG.
New to Contra
Sanchit is ready for their next project!
Validated Shannon, Keygraph’s (U.S.-based cybersecurity company) autonomous AI penetration tester, powered by the Anthropic Claude API, against a deliberately vulnerable FastAPI application using a proof-by-exploitation approach.
Demonstrated successful exploitation of SQL Injection, Authentication Bypass, NoSQL Injection, and Rate-Limit Abuse, with detailed technical findings and remediation guidance.
Authored a technical blog post and created a walkthrough video that earned 1,000+ views, showcasing Shannon’s real-world AI-driven hacking capabilities to developers and security teams.
1
38
Built an advanced conversational AI engine that gives AI assistants long-term memory, adaptive personality, and context-aware responses. The system remembers past conversations, retrieves relevant information automatically, and adjusts its communication style over time. Designed as the core intelligence layer for the MARVE AI Assistant platform, enabling more natural and personalized user interactions.
Project Highlights
Developed a persistent memory system that stores important conversations and recalls them when relevant.
Implemented Retrieval-Augmented Generation (RAG) for context-aware responses across multiple sessions.
Built a dynamic personality engine with configurable traits such as helpfulness, curiosity, wit, and motivation.
Designed a dual-model architecture that balances fast responses with deeper reasoning when additional context is needed.
Integrated semantic search using vector embeddings and Qdrant for efficient memory retrieval.
Added multi-session support so each user maintains isolated conversation history and preferences.
Engineered an asynchronous, production-ready architecture optimized for scalability and low-latency performance.
Key Deliverables
Conversational AI engine with long-term memory and personalized responses.
Vector-based knowledge retrieval system for remembering past conversations.
Production-ready Python backend with session management, analytics, and extensible architecture.
0
12
AI-powered cloud cost intelligence platform using Claude, MCP, AWS Cost Explorer, and LangGraph.--
Project Description--
Built an advanced AI-driven AWS cost analysis platform that allows users to query multi-account AWS spending in natural language using Claude. Developed a custom Model Context Protocol (MCP) server exposing AWS Cost Explorer and Amazon Bedrock usage data as tools to AI agents. Implemented support for local and remote MCP deployments over HTTPS, cross-account IAM role assumption, and an interactive Chainlit dashboard powered by LangGraph. Containerized the entire system with Docker for production-ready deployment.
Detailed Work Highlights --
1- Designed and built a custom MCP server in Python to expose AWS Cost Explorer and CloudWatch usage metrics to AI agents.
2- Integrated Anthropic Claude API (https://www.anthropic.com/api ) for natural-language querying of AWS cost and Bedrock usage data.
3- Enabled cross-account AWS cost visibility using IAM role assumption across multiple AWS accounts.
4- Implemented detailed cost breakdowns by service, region, instance type, and day.
5- Added Amazon Bedrock usage analytics by model, user, region, and hourly invocation patterns.
6- Built a LangGraph (https://www.langchain.com/langgraph ) agent capable of autonomously invoking MCP tools to answer cloud cost questions.
7- Developed an interactive Chainlit chat interface for real-time conversational cloud cost analysis.
8- Deployed the solution locally and remotely using Docker, EC2, SSE transport, and Nginx-based HTTPS reverse proxy.
9- Implemented secure multi-environment configuration using environment variables and AWS IAM best practices.
10- Authored comprehensive documentation and deployment instructions for production-ready setup.
Key Deliverables (Client Side)
1. AI assistant for answering cloud cost questions in plain English
2. Unified dashboard for AWS and AI infrastructure spending
3. Multi-account cost visibility across all AWS accounts
4. Detailed breakdown of costs by service, region, and time period
5. Secure cloud deployment with Docker and HTTPS
6. Documentation for setup, usage, and maintenance
1
24