I’m Md Ashik Adnan, a backend application engineer and AI engineer focused on building production-grade systems with Python (FastAPI), robust APIs, and reliable data pipelines.
I’m Md Ashik Adnan, a backend application engineer and AI engineer focused on building production-grade systems with Python (FastAPI), robust APIs, and reliable data pipelines. Over the years, I’ve delivered end-to-end applications—from microservices, caching, and background jobs to clean, maintainable database and deployment workflows on cloud infrastructure.
Since late 2023, I’ve deepened my work in applied AI: RAG and vector search (Qdrant/FAISS/Chroma/Pinecone/AstraDB), agentic orchestration (LangChain/LangGraph, CrewAI, AutoGen, Google ADK), and MLOps/CI/CD with Docker and AWS. I enjoy turning complex AI ideas into usable products, including financial research/planning systems, document-processing pipelines like Receipt OCR, and computer vision projects built with the YOLO family and OpenCV.
Applied OpenCV and detectors from the YOLO family (v5, v8, v11, v12, NAS) and the RF-DETR transformer architecture for drone imagery, vehicle detection, human activity recognition, PPE detection, and pothole detection. Published an interactive gallery with video and image demonstrations.
Portfolio link: https://www.twine.net/signin
GitHub link: https://www.twine.net/signin
Created 38 production-grade n8n workflows — customer support, lead capture, Telegram and WhatsApp AI assistants (OpenAI, DeepSeek R1, Whisper transcription), document summarization, news and finance agents, RAG with long-term memory (Supabase, Notion), and MCP integrations spanning Gmail, Google Sheets, Calendar, Contacts, Drive, Telegram, WhatsApp, and Notion. Strengthened data orchestration fundamentals with Apache Airflow (DAGs, scheduler, metadata database, batch orchestration) and Apache Kafka (topics, producers and consumers, offsets, brokers and clusters, retention). GitHub link: https://www.twine.net/signin | https://www.twine.net/signin | https://www.twine.net/signin
Architected a production-grade Digital Twin reference system — Next.js static frontend on S3 with CloudFront, serverless Python/FastAPI backend via API Gateway and AWS Lambda with S3-based conversation memory and stateless LLM pattern, LLM inference via Amazon Bedrock, Terraform infrastructure as code, and GitHub Actions CI/CD with multi-environment promotion (dev, staging, prod) and automated teardown for cost control. Containerized workloads with Docker. Also built Alex multi-agent financial planning SaaS and AI Cybersecurity Analyzer. GitHub link: https://www.twine.net/signin
Delivered multi-agent systems across four frameworks — CrewAI (financial researcher, stock picker, coding agent, automated SE team), LangGraph (web automator with push notifications, Sidekick co-worker), Microsoft AutoGen v0.4 (single agents with tool use and structured output, RoundRobinGroupChat teams with termination strategies and state persistence, human-in-the-loop, Swarm, GraphFlow, Docker execution), and Google Agent Development Kit (SequentialAgent, ParallelAgent, Session/State/Runner with Pydantic, persistent SQLite agent with DatabaseSessionService, callbacks with guardrails, LiteLLM routing, Tavily search). Integrated Model Context Protocol (MCP) with a custom MCP server and a Yahoo Finance MCP server with Streamlit visualization. Built database AI agents for natural-language CSV and SQL querying via LangChain agents, OpenAI function calling, and the Assistants API with SQLAlchemy, SQLite, and Streamlit. GitHub link: https://www.twine.net/signin | https://www.twine.net/signin | https://www.twine.net/signin | https://www.twine.net/signin
Built LLM applications across 19 project areas using LangChain — OpenAI GPT-4o and Ollama (gemma:2b) with LCEL chaining, document loaders (PyPDF, web, ArXiv, Wikipedia), text splitters (RecursiveCharacter/JSON), embeddings, vector stores (FAISS, Chroma, AstraDB), conversation-history chatbots, RAG document Q&A with Groq, SQL database agents, summarization, and CodeLlama and NVIDIA NIM integrations. Fine-tuned an end-to-end binary text classifier (food vs not-food image captions) with the Hugging Face ecosystem — custom dataset creation, tokenization, AutoModelForSequenceClassification, Trainer, Evaluate, Accelerate — published as a shareable Gradio demo on Hugging Face Spaces.
GitHub link: https://www.twine.net/signin | https://www.twine.net/signin
Completed a 10-notebook PyTorch program (tensors, workflow, classification, CNNs with FashionMNIST, custom datasets via TorchVision/ImageFolder, modular data_setup/engine/model_builder/train architecture, transfer learning, experiment tracking, Vision Transformer paper replication, deployment) and an 11-notebook TensorFlow program (regression, classification, CNNs, transfer learning with feature extraction, fine-tuning, and scaling using EfficientNet and ResNet50V2, TensorBoard tracking, Food Vision Big with 101 classes, NLP with word embeddings and Universal Sentence Encoder, SkimLit sequential sentence classification, BitPredict time series forecasting). Applied modular code architecture and experiment tracking throughout. GitHub link: https://www.twine.net/signin | https://www.twine.net/signin
Built 6-module Google ADK learning path (SequentialAgent, ParallelAgent, Session/State/Runner, persistent SQLite agent, agent/model callbacks with guardrails) and comprehensive AutoGen v0.4 program (single agents, RoundRobin teams, termination, state persistence, human-in-the-loop, Swarm, GraphFlow, Docker execution). Extended with CrewAI and LangGraph multi-agent projects and MCP-based stock researcher with Streamlit and Yahoo Finance integration. Stack: Google ADK, AutoGen, CrewAI, LangGraph, LiteLLM, MCP, Tavily, Streamlit, Docker.
GitHub link: https://www.twine.net/signin | https://www.twine.net/signin | https://www.twine.net/signin
All RAG-related work consolidated into one section, covering six GitHub repositories: rag (https://www.twine.net/signin — Production RAG pipelines from the 14-module curriculum: Restaurant Menu Automation (PDF menus → images → structured Excel via GPT-4o vision, PyMuPDF) and Smart Cookbook system (cookbook RAG over unstructured recipes, Phase 2 agentic RAG). Stack: LangChain, LangGraph, Qdrant, FAISS, Chroma, OpenAI, Gemini, Docling, Docker. rag_udemy (https://www.twine.net/signin — 14-module RAG curriculum: embeddings, vector stores, advanced chunking, hybrid search, query enhancement, multimodal RAG, agentic/autonomous/multi-agent RAG, CRAG, adaptive RAG, persistent memory, CAG, graph databases, LangGraph search app. Applied to restaurant menu automation, smart cookbook agentic RAG, multimodal financial research with Qdrant hybrid search, Docling, Qwen2.5-VL. Stack: LangChain, LangGraph, Qdrant, FAISS, Chroma, OpenAI, Gemini, Docling, Docker. knowledge_graph_rag (https://www.twine.net/signin — Hybrid retrieval over Neo4j knowledge graph with vector similarity + graph traversal, Cypher query generation, fulltext search, OpenAI embeddings, conversational memory. Einstein graph → healthcare CSV graph → production Roman Empire RAG with chat history. Stack: Neo4j, LangChain, LangChain-Neo4j, OpenAI, tiktoken, Wikipedia. multi_agent_deep_rag (https://www.twine.net/signin — Multi-agent and Deep RAG: agentic RAG using CrewAI, AutoGen, Microsoft Agent Framework; Deep RAG; production RAG patterns. MCP for Finance stock researcher (Streamlit), Docling multimodal extraction, Qdrant bulk ingestion, hybrid search + cross-encoder re-ranking, multimodal financial research agent, Deep Finance Researcher TODO planner. Stack: CrewAI, AutoGen, Microsoft Agent Framework, LangGraph, Qdrant, Docker, Docling. advance_agentic_rag (https://www.twine.net/signin — RAGWire: production-grade RAG toolkit for document ingestion and retrieval with hybrid search. Reusable Python library, eight agent implementations (LangChain, LangGraph, CrewAI, AutoGen, Microsoft Agent Framework), OpenAI-compatible FastAPI backend, Chainlit chat frontend, all backed by Qdrant. Stack: Qdrant, LangChain, LangGraph, CrewAI, AutoGen, Microsoft Agent Framework, FastAPI, Chainlit. private_agentic_rag (https://www.twine.net/signin — Private RAG with LangGraph v1 and Ollama: Corrective RAG, Reflexion, Self-RAG, Adaptive RAG, MySQL agent — local-first, optional cloud routing via OpenRouter. MySQL agent, PageRAG ingestion, retrieval and re-ranking, agentic PageRAG, MCP Airbnb hotel search. Stack: LangGraph, Ollama, ChromaDB, OpenRouter, MySQL, MCP.
Developed hybrid retrieval system combining vector similarity and graph traversal over Neo4j knowledge graph with Cypher query generation, fulltext search, and OpenAI embeddings. Progressed from simple Einstein graph to healthcare domain graph from CSV to production Roman Empire RAG with chat history and context-aware generation. Stack: Neo4j, LangChain, LangChain-Neo4j, OpenAI, tiktoken, Wikipedia. GitHub link: https://www.twine.net/signin
Built food vs not-food binary text classification model using PyTorch and Hugging Face Transformers, Datasets, Tokenizers, Evaluate, and Accelerate. Created custom caption dataset, configured AutoModelForSequenceClassification and TrainingArguments, trained with Trainer API, evaluated performance, and published interactive Gradio demo on Hugging Face Spaces. Demonstrates transfer learning and end-to-end NLP pipeline ownership. GitHub link: https://www.twine.net/signin
Deployed full-stack and AI applications demonstrating end-to-end delivery: Sales Analytics Dashboard (revenue, profit trend, product, regional heatmap, channel mix), Financial Dashboard (dual-axis KPI, profit margin, transaction tracking), Edenthought E-Commerce (Django, authentication, cart, product categorization), FinAlly AI Trading Workstation (market analysis), Kanban Studio (project management with kanban boards), Prelegal Legal Platform (case management, secure auth), and Receipt OCR (JPG/PNG upload to editable structured table with inline editing, currency switching, Markdown export, per-browser history). Stack: Django, React, Netlify, Render.
Portfolio link: https://www.twine.net/signin | https://www.twine.net/signin
Source URL: https://www.twine.net/signin | https://www.twine.net/signin
Continued full-time applied AI engineering work focusing on RAG, agentic systems, computer vision, and deployment/observability workflows, including agentic financial and document-processing applications.
Focused on Deep Learning, Generative AI, Agentic AI, RAG systems, and MLOps through structured study and applied project delivery. Built multiple public FastAPI and GenAI/LangChain/LangGraph repositories and live deployed applications, including production-style backend patterns and end-to-end RAG/agentic workflows with evaluation and vector-store ingestion.
Developed and maintained RESTful microservices for product center systems using Java concurrency (multi-threading), Spring Boot, Spring Cloud, and scheduled tasks. Implemented and enhanced frontend features using JavaScript, React, Redux, and CSS, integrating with MySQL-backed services. Delivered features iteratively within an agile environment and ensured reliability through unit and integration testing. Worked with CI/CD and AWS services (including EC2, S3, RDS, ECS, Fargate, ELB, Lambda, API Gateway, DynamoDB, CloudFormation, CodePipeline) as part of development and deployment workflows.
Post a job and invite Md, or contact them directly. No hidden commission.
