Building AI systems
that go beyond demos.
I am Swaraj Chattaraj, an AI Engineer specializing in agentic workflows, self-correcting RAG loops, and production-grade reasoning architectures. I bridge the gap between probabilistic models and deterministic software engineering.
Core Research &
Development Focus
I focus on building production-style AI tools. My work centers on self-correcting RAG systems, structured data outputs, and evidence-provenance metrics, bringing developer rigor to probabilistic AI workflows.
Designing multi-stage graph loops using LangGraph (Planner → Retriever → Synthesizer → Critic) that process queries like human analysts.
Engineering self-correcting loops where a Critic agent verifies claims against grounding materials, triggering supplementary searches to mitigate hallucinations.
Logging evidence attributes (relevance scores, paths, URLs) to generate citation markers that link directly back to verified grounding sources.
Using `asyncio` and `aiohttp` to parallelize multi-source web scraping and vector query executions, reducing search latencies under heavy query loads.
Education
Bachelor of Technology — Artificial Intelligence
Currently in my 3rd year (6th Semester). Studying core principles of deep learning architectures, classical optimization methods, and parallel computing infrastructures.
Skills Inventory
Flagship Development
ARIA: Autonomous Research Assistant
A 5-stage agentic RAG pipeline built with LangGraph. Instead of performing single-pass vector queries, ARIA plans query structures, retrieves text concurrently from multiple databases, evaluates claims, runs self-critique loops, and compiles a PDF report. Read the Installation Guide → for macOS & mobile setup.
RAGPro: Modular Document Indexing
A high-performance modular RAG pipeline focused on vector search indexing. Built to evaluate document chunking algorithms, test FAISS vector similarity search speeds, and benchmark response generation using OpenRouter LLM gateways.
Interested in research collaboration or testing agent setups?
I am always open to discussing stateful RAG loops, claim verification critique systems, or custom LLM integrations.
Get In TouchContact Channels
Submit a message through the console interface on the right. Form validation checks will run client-side before sending.