I'm Arun Addagatla, a Founding Engineer - AI at Lamatic.ai, focused on building robust AI platforms that move from prototype to production with speed and reliability.
- 🚀 Architecting systems across AI, backend, infrastructure, DevOps, and critical frontend components
- 🧠 Specialized in LLMs, Conversational AI, Agentic AI, Semantic RAG, and MLOps
- ⚙️ Passionate about high-scale inference, deployment orchestration, and enterprise integrations
- 🎓 Bachelor's in Computer Engineering (MCT's RGIT, University of Mumbai) with 9.6 CGPA
🧩 Content OS (in development)
AI-powered content orchestration monorepo that guides users through the entire content creation pipeline — clarify → outline → write → edit → publish to Blogger — with autonomous agents handling each stage.
A Model Context Protocol (MCP) server exposing 35+ tools over the Dev.to (Forem) API v1. Lets AI assistants like Claude and Cursor draft, edit, publish, and manage Dev.to content programmatically.
Supports stdio, Streamable HTTP, and Cloudflare Workers transports. Multi-arch Docker image available on GHCR.
End-to-end voice assistant supporting 10+ Indian languages including Hindi and Marathi. Built S2T with Conformer models on Triton, T2S with Fastpitch, and a RAG pipeline using LangChain with embedding + reranker models.
📄 PDFChat
Chat with any PDF — an LLM-powered assistant for grounded Q&A over uploaded documents.
March 2024 – Present · Miami, Florida, United States
- Built over 80% of Lamatic's core stack as the first engineering hire — spanning AI systems, backend infrastructure, and critical frontend components.
- Designed and trained an intelligent AI Agent that understands Lamatic's flow configurations and autonomously builds workflow pipelines from user requirements.
- Built a configurable AI evaluation framework with LLM-as-a-judge capabilities for automated quality scoring.
- Created an internal AI-powered Hiring Agent using multimodal agents (resume parsing + video response analysis), reducing recruiter workload by 70%.
- Architected a high-performance serverless executor processing 1M+ monthly requests and a deployment engine handling 1K+ deployments/min.
- Reduced deployment latency from 2 minutes to 15 seconds — an 87% improvement.
- Engineered secure (VPC) and scalable Kubernetes-based ETL pipelines for Drive, S3, and SharePoint with OAuth for Google, Microsoft, and GitHub.
- Developed advanced AI/data processing nodes including Semantic RAG and a Multi-Agent Supervisor.
- Built real-time webhook integration systems for Slack and Teams for instant event-driven sync.
- Designed Lamatic's version control system (VCS) with native GitHub integration for automatic flow sync.
October 2022 – March 2024 · Mumbai, India
- Designed and deployed conversational chatbots and voicebots handling 95% of customer queries using word embeddings, reranking, and instruction-tuned Zephyr and GPT-4.
- Fine-tuned open-source LLMs (Mistral, LLaMA-2) with LoRA/PEFT on bilingual datasets, improving fluency by 30%.
- Built a scalable LLM Inference Engine with dynamic batching and multi-GPU support, achieving 106 tokens/sec throughput.
- Optimized Whisper V3 large ASR models with ONNX/TensorRT and Triton deployment, reducing latency to 0.1–0.4s.
- Implemented multimodal search using LLaVA and GPT-4, boosting search performance by 60%.
September 2021 – August 2022 · Bangalore, India
- Built a serverless multilingual sentiment analysis system on AWS (Lambda, ECR), reducing processing time by 50%.
- Designed CI/CD pipelines and Docker containerization with integration and load testing suites.
- Enhanced NER and text classification models by 30% through hyperparameter tuning.
- Implemented an anomaly detection system using CloudWatch, Prometheus, and Grafana — reducing false positives by 60%.
July 2020 – September 2020
- Developed an ML model, converted it to ONNX, and deployed it with a Golang runtime for client-side productivity tracking.
- Why LLMs Need Memory — Lamatic.ai Community Session (Mar 2026) · YouTube
- Why Prompting Isn't Enough: The Case for RAG — Lamatic.ai Community Session (Jan 2026)
- What is MCP and How It Works — Daytona Developers Club Tour '25, Mumbai (May 2025)
Active member of global AI and developer communities — Maxpool · Entrepreneurs Arch · Langfuse · Cloudflare Community · Learn AI Together · AG2 · DAIR.AI
MCT's Rajiv Gandhi Institute of Technology — University of Mumbai
Bachelor of Engineering in Computer Science · CGPA: 9.6/10
Coursework: OS, Data Structures & Algorithms, ML, Networking, NLP, Software Engineering, DBMS, Computer Networks, Probability & Statistics.
- Neural Networks and Deep Learning — deeplearning.ai (Coursera)
- Deep Learning A-Z: Hands-On Artificial Neural Networks — Udemy
- The Data Science Course 2020: Complete Data Science Bootcamp — Udemy
- Master Python Programming — Udemy
- Tools for Data Science — IBM (Coursera)
- Foundation of Data Science — IBM (Coursera)
- Building reliable GenAI + Agentic AI systems for enterprise use-cases
- Scalable deployment and inference platforms
- Robust MLOps and production AI infrastructure
- Retrieval-augmented systems, MCP integrations, and autonomous workflow automation