Adarsh Kumar Singh Saviour5538

`$ whoami`

class Saviour5538:
    name        = "Saviour"
    location    = "India 🇮🇳"
    focus       = ["Agentic AI", "RAG & Graph RAG", "Deep Learning", "AI Safety"]
    stack       = ["Python", "C++", "SQL", "LangChain", "LangGraph", "MCP", "A2A"]
    building    = "Production-grade AI systems that think, plan, and act autonomously"
    learning    = ["Graph RAG", "A2A Protocol", "MCP Servers", "Constitutional AI"]
    motto       = "Don't just use AI — build systems that use AI to build more AI 🔄"

`$ cat tech_stack.json`

🐍 Languages

🧠 Machine Learning & Deep Learning

🤖 Agentic AI & LLM Frameworks

🔍 RAG · Graph RAG · Vector Databases

🗄️ Databases

🔧 Backend & MLOps

`$ ls -la ./flagship-projects/`

🛡️ Trustworthy AI — Guardrails

Production-grade AI safety system wrapping LLaMA 3.3 70B with a 9-step guardrail pipeline — injection detection, toxicity classification, PII redaction, RAG hallucination checking & Constitutional AI self-critique.

LangChain RAG Qdrant Groq Streamlit

🕸️ A2A Protocol

Full implementation of Google's Agent-to-Agent protocol. Autonomous AI agents that discover each other, exchange tasks & compose results over HTTP using JSON-RPC 2.0. Includes web dashboard & Groq API.

A2A FastAPI Groq Python

🏭 Agentic CRUD

Multi-agent pipeline that autonomously builds a full Streamlit + SQLite app from plain English. BA, Architect, PM, Dev & QA agents collaborate end-to-end — zero manual coding required.

LangGraph LangChain Groq SQL

🧬 SemanticDNA

Bio-inspired semantic search without vector embeddings — pure PostgreSQL. A novel approach using biological DNA analogy for efficient similarity search at scale — an alternative to vector databases.

PostgreSQL SQL Graph RAG Python

🔬 Agentic Realtime Researcher

Production-grade Agentic RAG with autonomous query rewriting, local FlashRank reranking & persistent memory for grounded, real-time web research.

LangChain RAG Vector DB MCP

📝 Text Summarization — Transfer Learning

Compares LoRA, Prefix Tuning & P-Tuning for summarization on T5-small. Evaluated with ROUGE & BERTScore — efficient fine-tuning with minimal trainable parameters.

PyTorch HuggingFace LoRA Deep Learning

`$ git log --stat`

`$ cat contribution_graph.txt`

`$ ping connect`

> Building agents that build agents.
> Every commit is a step toward autonomous intelligence.

Navigation Menu