— AI SOFTWARE ENGINEER —
Multi-Agent Swarms · Production RAG · Enterprise Guardrails
WHO AM I
I'm Adarsh Kumar Singh, a B.Tech student who decided to stop watching tutorials and start building real, deployable AI systems from scratch.
Over 14 relentless weeks, I went from Python fundamentals to engineering autonomous multi-agent swarms that fix CI/CD pipelines without human intervention. I didn't follow a bootcamp. I didn't copy boilerplate. I built every line of code, every architecture, every guardrail myself.
The result? A production-grade portfolio that proves I can ship AI systems that solve real, painful engineering problems.
ReAct Loops, Tool Calling, Guardrails, RAG Pipelines
FastAPI, SSE Streaming, REST APIs, Pytest
Prompt Injection Prevention, Scope Guardrails, Model Failover
THE JOURNEY
From print("hello") to autonomous AI agents that write, test, and deploy code.
Mastered Python, NumPy, and Pandas. Built data processing pipelines from scratch.
Built a Document Risk Prediction system. Feature engineering, hyperparameter tuning, model comparison.
Built database-backed REST APIs using FastAPI. Pydantic validation and production deployment patterns.
First contact with Large Language Models. Built an LLM Assistant with structured prompting and JSON output.
Full Retrieval-Augmented Generation pipeline with ChromaDB, semantic chunking, reranking, and hallucination control.
Built multimodal agents with Function Calling. Agents that read files, scrape websites, and execute code.
Real-time Eval Dashboard with SSE streaming, LLM-as-a-Judge, and automatic Model Failover (70B → 8B).
Hardened AI agent with Input/Output guardrails blocking prompt injection and out-of-scope queries.
Automated GitHub PR reviewer and a Meeting Intelligence pipeline extracting decisions and action items.
The capstone. A multi-agent system that intercepts CI/CD failures, diagnoses bugs, writes patches, runs tests, and opens Pull Requests — all without a human.
THE PROOF
Not tutorials. Not clones. Real systems solving real problems.
A low-latency, bidirectional voice agent utilizing WebSockets and Server-Sent Events (SSE) for real-time speech-to-text (STT) and text-to-speech (TTS) streaming. Optimized for sub-500ms time-to-first-byte (TTFB).
A multi-agent orchestration framework that intercepts CI/CD pipeline failures, diagnoses root causes, writes code patches, verifies fixes via sandboxed testing, and opens Pull Requests — fully autonomously.
A hardened compliance AI agent with Input/Output Guardrails, persistent memory, structured logging, and an LLM-as-a-Judge Evaluation Dashboard with automatic Model Failover routing.
Upload any PDF. Ask questions. Get cited, accurate answers. Built with ChromaDB vector search, semantic chunking, reranking, and hallucination control.
Submit any GitHub PR URL. Get a structured LLM-powered code review with bugs, security issues, severity ratings, and approval decisions. Connects to real GitHub diffs.
MY ARSENAL
LET'S CONNECT
I'm actively seeking AI Engineer and Backend Software Engineer roles.
BUILDING IN PUBLIC
Real-time stats pulled from the GitHub API.
EXPLORE
Type help to see available commands.