Hi, I'm Souvik Chakraborty — Fullstack & AI Engineer

Architected an enterprise Multi-Agent platform using LangGraph (24-node cyclical StateGraph) and FastAPI, building a dual-tier supervisor router (regex fast-path + forced LLM tool-calling) to orchestrate complex data, analytics, simulation pipelines while cutting query routing latency by 40%.

Engineered autonomous ReAct tool loops (up to 5 iterations) and MCP transports to integrate Datahub schema discovery and analytics microservices, enforcing thread-safe context propagation across concurrent workflows.

Built a resilient LLM Gateway and 3-tier persistence architecture using LiteLLM and AWS Bedrock (Claude 3.5 Sonnet / Haiku fallback), integrating Redis hot-caching, DynamoDB event-sourcing and SHA-256 prompt hashing to reduce token usage by 35% and latency by 40%.

Implemented Human-in-the-Loop (HITL) interactive planning gates with non-blocking WebSocket status streaming and dual-tier JSON/regex output parsing, rendering real-time dynamic ECharts visualizations with 99.9% parse reliability.

Modernized to a NextJS frontend and FastAPI backend architecture for a B2B SaaS threat modeling platform, designing database architecture for structured and JSON data to cut latency by 95% and boosting client engagement by 4x.

Built an SSE-based MCP server for Agentic capabilities, automated codebase vulnerability analysis and PR resolving, performed vulnerability assessment enforcing security controls and automating CI/CD deployments via Docker and AWS.

Engineered an end-to-end RAG pipeline over a threat database, implemented a Role-Based Access Control (RBAC) organization dashboard from scratch with Redis caching, integrating Jira APIs for threat-to-task automation and Razorpay for payment processing.

Executed comprehensive penetration testing, red-team simulations, and vulnerability assessments, enforcing OAuth2/JWT security controls and automating zero-downtime Docker/AWS production deployments.