Available for work

Hi, I'm Babitdor Kayang Khonglah

AI/ML Engineer specializing in LLM orchestration, RAG architectures, and building intelligent systems that scale.

Erlangen, Germany

About Me

Passionate about building intelligent systems and pushing the boundaries of AI

I'm an AI/ML Engineer specializing in LLM orchestration and RAG architectures. Currently pursuing my Master's in Artificial Intelligence at FAU Erlangen-Nürnberg, I build intelligent systems that leverage cutting-edge AI technologies.

My expertise spans multi-agent workflows, vector databases, and deploying AI solutions at scale. I'm passionate about building tools that enhance developer productivity and automate complex workflows.

5+
Projects Completed
3+
Years Experience
10+
Technologies

Education

M.Sc. Artificial Intelligence

Friedrich-Alexander-Universität Erlangen–Nürnberg

03/2024 – Present

B.Tech. Computer Science & Engineering

National Institute of Technology, Meghalaya

04/2018 – 04/2022

Technical Skills

Technologies and tools I use to build amazing things

Python

Language

JavaScript

Language

TypeScript

Language

C++

Language

LangGraph

AI/ML

LangChain

AI/ML

CrewAI

AI/ML

RAG

AI/ML

LLM

AI/ML

Docker

DevOps

PostgreSQL

Database

MongoDB

Database

Vector DB

Database

Git

DevOps

React

Frontend

Node.js

Backend

Projects

A selection of my recent work and experiments

Multi-Agent Research Intelligence System

Advanced AI-powered multi-agent research system orchestrating specialized AI agents to conduct comprehensive research, synthesize information, and generate professional reports. Features 10 research strategies, parallel swarm research, self-critique quality control, RAG knowledge management, and dual-format report generation (Markdown + PDF).

LangGraphLangChainPythonRAGChromaDBMulti-AgentOllama

SysML v2 Multi-Agent Workflow Automation

Sophisticated AI-powered system for generating and validating SysML v2 code using multi-agent architecture with RAG and vector database integration. Features intelligent code generation with context awareness, automated multi-step validation with syntax and semantic checks, automatic error detection and fixing, knowledge management with RAG system, solution memory using ChromaDB vector database, optional human approval with feedback integration, and real-time analytics dashboard.

LangGraphLangChainPythonRAGChromaDBOllamaStreamlitMulti-AgentJupyterPyTorch

Nami-Code: Deep Agent Framework & AI Coding Assistant

Open-source AI agent framework and terminal-based coding assistant enabling LLMs to handle complex multi-step tasks through advanced planning, context management, and parallel execution. Features planning tools, filesystem operations, subagent spawning, persistent memory, and MCP support for extensible tool integration.

LangGraphLangChainPythonPrompt EngineeringOpenAIAnthropicOllamaMCP

S.A.R.A.H. – Smart Assistant Real At Heart

A fully local desktop voice assistant powered by a fine-tuned LLaMA 3.2 model. Uses a real-time STT → LLM → TTS pipeline with LiveKit for voice streaming, Deepgram for speech-to-text, and Cartesia AI for natural voice responses. Can understand natural language commands and invoke tools like launching apps, reading emails, or organizing files — all while running 100% offline for complete privacy.

LLaMA 3.2LiveKitDeepgramCartesia AIVoice AssistantSTTTTSLocal AI

Speaker Voice Separation with Dual-Path Transformers

Enhancing the Dual-Path RNN framework by replacing the original recurrent modules with intra-chunk and inter-chunk Transformer layers. This hybrid architecture aims to better capture both local and global audio dependencies for superior multi-speaker separation. Intra-chunk Transformer applies self-attention within small audio chunks to model fine-grained, short-term temporal features, while Inter-chunk Transformer captures long-range context by attending across chunk sequences.

PyTorchTransformersAudio ProcessingSpeaker SeparationDeep LearningSI-SNRSDR

Assistive Sidewalk Segmentation: Fine-Tuning SAM 2.1 with a Custom Dataset for the Visually Impaired

Developed a comprehensive training and deployment pipeline for fine-tuning Segment Anything Model (SAM) 2.1 by Meta on custom sidewalk imagery, aimed at enabling assistive vision systems for the visually impaired. The project spans data preprocessing, dynamic prompt generation, mixed-precision training, checkpointing, and model evaluation, culminating in an interactive Streamlit application for real-time segmentation and inference.

SAM 2.1PyTorchHugging FaceStreamlitComputer VisionFine-TuningAssistive TechnologyFP16/BF16

ogAI: Multi-Model LLM Discord Assistant

Built a fully customizable AI-powered Discord bot that integrates multiple local and cloud-based language models (e.g., Ollama, Gemini, GPT-4) to deliver intelligent, context-aware responses across diverse use cases including general Q&A, summarization, and creative tasks. Features integrated multiple backends, flexible routing logic to dynamically select models based on task or user input, prompt templates, system messages, and conversation context tracking.

discord.pyOpenAI GPTGoogle GeminiOllamaLLMDiscord BotMulti-ModelAsync

N8N-Framework: AI-Powered SysMLv2 Systems Engineering Platform

Advanced AI-powered systems engineering platform integrating SysMLv2 with multi-agent LLMs to assist in designing, modeling, analyzing, and optimizing complex systems. Built on n8n workflow automation, features AI-driven code generation, multi-agent analysis (KPI-Analyst, SysML-Expert, MA-Solver), rigorous ANTLR4-based validation, Eclipse SysON integration for visual modeling, and comprehensive knowledge management with RAG capabilities.

n8nNext.jsLangGraphLangChainFastAPISysMLv2Multi-AgentRAGQdrantPostgreSQLDockerEclipse SysON

Get In Touch

I'm always open to discussing new opportunities, research collaborations, or interesting AI/ML projects.