MD Emdadul Haque

• AI & Generative AI

  LLM applications, retrieval and evaluation.

  • LLMs8.5 out of 10
  • Generative AI8.5 out of 10
  • RAG8.5 out of 10
  • Embeddings8.5 out of 10
  • Semantic search8.5 out of 10
  • Prompt engineering8.5 out of 10
  • Vector search8.0 out of 10
  • NLP8.0 out of 10
  • Tool / function calling8.0 out of 10
  • AI evaluation7.0 out of 10
  • RAG evaluation7.0 out of 10

• Agents, Local LLMs & Research

  Agentic systems, open-weight models and exploratory work.

  • AI agents7.5 out of 10
  • Ollama7.5 out of 10
  • MCP7.0 out of 10
  • Agentic AI7.0 out of 10
  • vLLM7.0 out of 10
  • Hugging Face7.0 out of 10
  • Computer vision6.5 out of 10
  • Knowledge graphs6.5 out of 10
  • GraphRAG6.0 out of 10
  • Fine-tuning / QLoRA5.5 out of 10
  • LangGraph5.5 out of 10
  • LlamaIndex5.0 out of 10
  • A2A4.5 out of 10

• Backend

  Enterprise .NET and Python service engineering.

  • C#9.0 out of 10
  • .NET Core9.0 out of 10
  • ASP.NET Core9.0 out of 10
  • Web API / REST APIs9.0 out of 10
  • Entity Framework8.5 out of 10
  • Microservices8.5 out of 10
  • Software architecture8.5 out of 10
  • Python8.5 out of 10
  • FastAPI8.0 out of 10
  • Async Python7.5 out of 10
  • Pydantic7.5 out of 10
  • Flask7.0 out of 10

• Databases & Search

  Relational, vector, search and cache stores.

  • SQL Server9.0 out of 10
  • PostgreSQL8.5 out of 10
  • ChromaDB8.0 out of 10
  • Qdrant7.5 out of 10
  • Redis7.5 out of 10
  • SQLite7.5 out of 10
  • Elasticsearch7.5 out of 10
  • Meilisearch7.0 out of 10
  • Neo4j6.5 out of 10

• Cloud & DevOps

  Azure-first delivery, containers and pipelines.

  • Microsoft Azure8.5 out of 10
  • Azure OpenAI8.5 out of 10
  • Azure App Service8.5 out of 10
  • Docker8.5 out of 10
  • Azure DevOps8.0 out of 10
  • Azure Functions8.0 out of 10
  • Kubernetes8.0 out of 10
  • CI/CD8.0 out of 10
  • Azure AI / Foundry6.5 out of 10
  • GitHub Actions5.5 out of 10
  • AWS4.0 out of 10
  • GCP4.0 out of 10

• Industrial / IoT

  Field hardware, protocols and edge inference.

  • Industrial IoT7.0 out of 10
  • Modbus RTU7.0 out of 10
  • RS4857.0 out of 10
  • MQTT7.0 out of 10
  • UART7.0 out of 10
  • ESP327.0 out of 10
  • M5Stack / StampPLC6.5 out of 10
  • Edge AI6.0 out of 10
  • ONNX Runtime6.0 out of 10
  • Honeywell (building automation)5.5 out of 10

1. Senior AI & Software Engineer (L-1)

   Akij iBOS Limited

   March 2021–toPresent

   Building and running enterprise software for the iBOS ERP ecosystem, and adding LLM, RAG and automation capabilities on top of it.

   users on supported systems

   ~2,000

   chatbot response (from ~40–50s)

   <5s

   engineers led (direct + indirect)

   4 + ~15

   • Design and build backend services and REST APIs in C#, ASP.NET Core / MVC, Web API and Entity Framework for ERP, supply chain (RTM), garments (RMG) and accounting products.
   • Support production systems used by approximately 2,000 people; work covers system analysis, system design, architecture, design patterns and database-driven application design.
   • Reduced the response time of an intelligent chatbot workflow from roughly 40–50 seconds to under 5 seconds.
   • Integrate LLMs (OpenAI, Azure OpenAI, open-weight models) into enterprise workflows: RAG pipelines, embeddings, vector databases, semantic search, tool calling and AI agents.
   • Build Python and FastAPI services alongside .NET systems, with asynchronous processing and background services for long-running work.
   • Deploy on Azure App Service and Azure Functions, using Docker, Kubernetes and Azure DevOps CI/CD pipelines.
   • Work on Industrial IoT and computer-vision solutions, including PLC, Modbus RTU / RS485 and MQTT integration with enterprise software.
   • Lead four engineers directly and coordinate around fifteen more indirectly.

   • C#
   • .NET Core
   • ASP.NET Core
   • Entity Framework
   • Python
   • FastAPI
   • Azure OpenAI
   • RAG
   • Embeddings
   • Vector databases
   • SQL Server
   • PostgreSQL
   • Azure
   • Docker
   • Kubernetes
   • Azure DevOps
   • Industrial IoT

• AI & RAG evaluation

  Studying evaluation for LLM and RAG systems: relevance, groundedness, fluency, retrieval quality and context quality, including Azure AI evaluation concepts.

  • RAG evaluation
  • Azure AI
  • Groundedness

• Azure AI Foundry & Foundry Local

  Working through model deployment on Azure AI Foundry, Foundry Local, and the regional availability and quota constraints that come with them.

  • Azure AI Foundry
  • Model deployment

• Agent frameworks & protocols

  Project-level exploration of LangGraph, LlamaIndex and A2A alongside MCP, mainly through the AgentMesh reference architecture.

  • LangGraph
  • LlamaIndex
  • A2A
  • MCP

• Fine-tuning & small models

  Experiments with QLoRA, Hugging Face tooling and open-weight or small language models. Research and experimentation, not production training.

  • QLoRA
  • Hugging Face
  • Open-weight LLMs

• Local LLM serving

  Running open-weight models with Ollama and vLLM on GPU hardware to evaluate enterprise AI that keeps data off external APIs.

  • Ollama
  • vLLM
  • Qwen
  • DeepSeek

Showing 9 of 9 projects

• AI & RAG

  MVP

  Enterprise RAG Platform

  Enterprise knowledge retrieval that grounds LLM answers in indexed company documents using embeddings, a vector database and embedding-based semantic search.

  Problem

  Enterprise knowledge is scattered across documents and systems, and keyword search returns file lists rather than answers. General-purpose LLMs do not know internal content and cannot be trusted to guess it.

  Solution

  Ingest documents, split them into chunks, embed each chunk, and store the vectors in a vector database. At query time, run embedding-based semantic search (vector similarity, such as cosine similarity) and optionally hybrid retrieval, assemble the retrieved context, and have the LLM generate an answer grounded in that context.

  Architecture

  Clean separation between ingestion, retrieval and generation so each stage can be evaluated and replaced independently.

  • Python
  • FastAPI
  • LLM
  • Azure OpenAI
  • Embeddings
  • +4

  View details

• AI Agents

  Prototype

  AI Interview System

  A multi-agent interview platform design covering interview planning, question generation, answer evaluation, behavioral analysis and report generation.

  Problem

  Structured interviews are time-consuming to prepare, hard to score consistently, and difficult to summarise into a comparable report.

  Solution

  Split the process across cooperating agents: a planner defines the interview structure, a question generator tailors questions to the role and profile, an evaluator scores answers, a behavioral analyzer processes visual signals, and a report generator produces the final assessment. Conversational and avatar interfaces are being explored for the candidate experience.

  Architecture

  One agent per responsibility, coordinated through a FastAPI backend.

  • Python
  • FastAPI
  • LLM
  • AI agents
  • RAG
  • +4

  View details

• Enterprise Systems

  Concept

  Enterprise Anomaly Detection Platform (EDAP)

  An architecture for collecting events across business systems, detecting anomalies, scoring risk and recommending controlled actions.

  Problem

  Irregularities such as unusual sales-order patterns are spread across many systems and are usually noticed late, if at all, because no single system sees the whole picture.

  Solution

  Collect events from each source into a common shape, run detection over the normalised stream, score the risk of each finding, alert the right people and recommend a next step, with any automated action kept under control.

  Architecture

  Business events (for example "sales order approved") are modelled separately from raw database changes.

  • Python
  • .NET
  • SQL Server
  • PostgreSQL
  • REST APIs
  • +1

  View details

• AI & RAG

  MVP

  AI Job Matching Engine

  Job recommendation and candidate matching that combines parsed profile and job features, semantic embeddings and a 100-point scoring framework.

  Problem

  Keyword matching misses candidates whose skills are phrased differently from a job description, and opaque rankings are hard to trust or tune.

  Solution

  Extract structured features from profiles and jobs, embed skills and text for semantic similarity, retrieve candidates or jobs by vector search, and rank them with a transparent 100-point recommendation framework.

  Architecture

  A 100-point scoring framework across relevant experience, duration, industry, seniority, latest category, job title, skills and education.

  • Embeddings
  • Semantic search
  • Vector database
  • Sentence Transformers
  • Backend APIs
  • +1

  View details

• Industrial AI

  Prototype

  Edge Industrial Vision & Telemetry

  Production-line monitoring that combines computer vision, edge inference and telemetry, with industrial control over Modbus RTU / RS485.

  Problem

  Manual counting and inspection on a production line is slow and error-prone, and the resulting data rarely reaches enterprise systems in a usable form.

  Solution

  Run vision models at the edge to detect and count events, publish telemetry over lightweight protocols, and integrate with backend systems. Control hardware is driven through Modbus RTU over RS485.

  Architecture

  Target applications: cement bag counting, defect detection, production monitoring and industrial event detection.

  • Computer vision
  • Edge AI
  • Python
  • MQTT
  • Modbus RTU
  • +4

  View details

• AI Agents

  Concept

  ERP MCP Server

  A concept for exposing ERP capabilities to LLM agents as governed tools through the Model Context Protocol.

  Problem

  Connecting LLMs to ERP systems ad hoc leads to fragile integrations and unclear access boundaries, and giving an agent direct database access is unacceptable in an enterprise.

  Solution

  Wrap selected ERP operations as MCP tools behind an authenticated server. The agent discovers and calls tools; the server enforces permissions, calls existing enterprise APIs and returns structured results.

  Architecture

  Existing ERP APIs remain the system of record; the MCP layer is an adapter, not a rewrite.

  • MCP
  • LLM
  • Tool calling
  • .NET
  • ASP.NET Core
  • +2

  View details

• AI Agents

  Reference

  AgentMesh

  A reference architecture for an enterprise AI platform: provider-agnostic LLM access, agent runtime, RAG and MCP behind a hexagonal core.

  Problem

  AI prototypes often hard-wire one LLM provider, one vector store and one orchestration library, which makes them difficult to evolve, test or move between clouds.

  Solution

  Use Clean / Hexagonal architecture so LLM providers, agent implementations, RAG providers, tools and infrastructure adapters are all replaceable behind ports.

  Architecture

  Ports and adapters for LLM providers: OpenAI, Azure OpenAI and local models.

  • Python
  • FastAPI
  • Pydantic
  • Async Python
  • LangGraph
  • +7

  View details

• AI & RAG

  Prototype

  Enterprise Knowledge Graph & Intelligence Portal

  Connects email history, ERP master data and Google Workspace activity in a knowledge graph to power AI search and relationship analysis.

  Problem

  Important context lives in email, ERP master data and workspace activity that are never joined, so relationship questions cannot be answered from any one system.

  Solution

  Integrate the sources into PostgreSQL and a Neo4j graph, then apply RAG and AI search over the connected data.

  Architecture

  Relational storage in PostgreSQL alongside a Neo4j graph of entities and relationships.

  • PostgreSQL
  • Neo4j
  • RAG
  • Knowledge graphs
  • Google Workspace APIs

  View details

• Research

  Research

  Enterprise Local LLM & Tool-Calling Platform

  Research into serving open-weight models locally with Ollama and vLLM so enterprise AI does not have to send all data to external APIs.

  Problem

  Some enterprise data should not leave the organisation, which rules out sending every request to an external LLM API.

  Solution

  Serve open-weight models (Qwen, DeepSeek) on local GPUs through Ollama and vLLM, and give them tool-calling capabilities for agent workflows.

  Architecture

  Hardware experimentation on an RTX 4090 and a dual-GPU server environment.

  • Ollama
  • vLLM
  • Qwen
  • DeepSeek
  • Hugging Face
  • +1

  View details