Md. Shihabuddin Sadi

AI / RAG Application Developer · Production chatbots and AI agents that ship · Multilingual support, grounded retrieval, no hallucinations · Ex-Samsung R&D · 15+ years of software engineering

View My GitHub Profile

I build production RAG chatbots and AI agents that ship.

Multilingual support. Grounded retrieval. No hallucinations. Built for real users, real traffic, real outcomes.

Ex-Samsung R&D · 15+ years of software engineering · Based in Dhaka, Bangladesh · Available worldwide remote

📅 Book a 30-min call See my work


What I Build

Production RAG chatbots over your docs, PDFs, Notion, or SQL — with citations, not hallucinations. Multilingual AI agents that handle Bangla, Banglish, English, and other low-resource or script-mixed languages, which makes them especially useful for South Asian, Middle East, and emerging-market audiences.

Beyond that, I build Messenger, WhatsApp, Telegram, and Slack bots wired to real business data, custom AI copilots embedded inside SaaS products, and the evaluation pipelines, observability, and guardrails that keep all of it from silently regressing in production.

I also handle the cloud infrastructure to keep it running reliably — Kubernetes, AWS, Terraform, CI/CD, Prometheus, Grafana. One contractor, one accountable line, no hand-off between the AI person and the DevOps person.


Featured Project — Minimal RAG Chatbot

A production multilingual RAG chatbot deployed on Facebook Messenger for Minimal Limited, an interior design company in Dhaka. Customers send questions in Bangla, Banglish, or English — the bot always replies in formal Bangla, grounded in a curated knowledge base, with graceful human takeover when confidence is low.

The one decision that paid off most: embed the question, not the answer. Customers send questions, so questions belong in the searchable space. Fixed more “wrong answer” bugs than any prompt tweak.

Key decisions:

Stack: Python 3.13 · OpenAI (text-embedding-3-small, gpt-4o-mini) · FAISS (IndexFlatIP, L2-normalized) · FastAPI · Uvicorn · Facebook Graph API · Pytest

At a glance: 224 curated Q&A entries · 14 intents · top-k=3 retrieval · embedding-dim 1536

Read the full case study · View on GitHub


Why Teams Hire Me

I’ve shipped real software for 15+ years — not just AI demos.

I bring engineering rigor: evals, logging, retrieval tuning, and guardrails. The unglamorous work that decides whether your AI survives contact with real users.

And because I can build both the AI and the cloud infrastructure it runs on, there’s no hand-off between the AI person and the DevOps person. One contractor, one accountable line.


How We’d Work Together

  1. 30-min discovery call — tell me about your product, your data, and where AI fits
  2. Scoped proposal within 48 hours — what I’d build, timeline, cost
  3. Build, ship, iterate — typically 2–6 weeks for a production RAG pilot
  4. Optional ongoing support — evals, observability, infra, and iteration
📅 Book a discovery call


What People Say

“Sadi is a highly skilled solutions architect, DevOps expert, and technical project manager with a deep understanding of software development, system architecture, and cloud infrastructure. His ability to streamline complex processes, optimize workflows, and enhance system efficiency made him a key asset to Samsung’s R&D initiatives. I highly recommend Md. Shihabuddin Sadi to anyone seeking a dedicated, skilled, and forward-thinking technical leader.”

— Md Elme Focruzaman Razi, Senior Staff Engineer at Samsung R&D Institute Bangladesh (LinkedIn)


Tech Stack

AI / LLM / RAG


Programming


DevOps & Cloud Infrastructure


Monitoring & Observability


Other Work

A selection of supporting projects across cloud infrastructure, DevOps automation, and software engineering.

Single-Node Kubernetes Cluster Multi-service web app (React, Node.js, MongoDB) deployed on a single-node Kubernetes cluster using Minikube — Deployments, Services, Ingress, ConfigMaps, Secrets, PV/PVC. Tech: Kubernetes · Docker · Minikube · NGINX Ingress

Kubernetes on AWS (EKS) End-to-end CI/CD on AWS EKS — Fargate, eksctl, Jenkins, DockerHub, and ECR integrations. Tech: AWS · EKS · Fargate · Jenkins · Docker

Terraform IaC Infrastructure as Code patterns for repeatable, auditable cloud deployments. Tech: Terraform · AWS

Prometheus + Grafana Monitoring Monitoring and observability setup for cloud-native applications. Tech: Prometheus · Grafana

Ansible Automation Configuration management and infrastructure automation playbooks. Tech: Ansible · Playbooks

Python Automation Engineering utilities and workflow automation in Python. Tech: Python · Bash

See all repositories on GitHub


Background

15+ years of software engineering across embedded systems, mobile, full-stack, cloud, and AI.

I started at Samsung R&D Bangladesh, where I worked on firmware for handsets shipped across Middle East, Africa, and Bangladesh — including the Bengali Calendar for the Bangladesh region and language support for Swahili, Yoruba, Igbo, Hausa, and Amharic on Samsung feature phones used by millions. That’s where multilingual production software became muscle memory, which weirdly turned out to be great prep for the multilingual RAG work I do now.

After Samsung, I co-founded Training Pool, Bangladesh’s first online training marketplace and SaaS platform. Took it from idea to live product with paying users. Before that, I ran a small dev studio building Android multiplayer games and Bangladesh client projects.

These days I run Grounded Labs, shipping production RAG applications and AI agents for founders, agencies, and mid-market teams.

See full work history on LinkedIn


Blog

Read my latest posts · RSS feed


Let’s Talk

If your chatbot is hallucinating, your AI feature isn’t making it past the demo stage, or you want to add a real RAG system to your product without it embarrassing you in front of customers — let’s talk.

📅 Book a 30-min call ✉️ Email me