Engineering journal · Business problems & practical delivery
How I Build RAG Systems and Agent-Based Applications
The AutoBot: connecting vector search and knowledge graphs
I worked on The AutoBot as a full-stack AI engineer. The assistant combines vector search and knowledge graphs in a GraphRAG application. My project record identifies GraphRAG and Neo4j as part of that work. The business focus is retrieving information for an enterprise assistant.
The implemented capability is an assistant that uses both vector-based retrieval and graph-based knowledge. This is a concrete example of my RAG architecture experience: connecting retrieval components to an application that users can interact with. It also connects my backend and AI work within the same product.
Pinecone and LangChain: improving semantic retrieval
At Cybernef Technologies, I designed a RAG system with Pinecone and LangChain to improve the accuracy of semantic data queries. This was information-retrieval work: bringing relevant material into the interaction so the application could respond using the underlying information.
I also worked on pytest coverage for RAG retrieval pipelines and configured observability tools. The optimization was therefore supported by engineering around the retrieval behavior, rather than being limited to prompt changes. My résumé records an improvement in semantic retrieval accuracy without a published numerical benchmark.
Planet AI and Twin Companion: two different AI workflows
On Planet AI, I worked on a full-stack workflow builder using multi-agent orchestration to automate business tasks. On Twin Companion, my role covered the frontend and AI integration for a conversational application with session memory and adaptive sentiment matching.
Planet AI demonstrates my work on coordinating AI-driven tasks. Twin Companion demonstrates conversational behavior and continuity. Together, they show why my agent architecture and application work spans different product needs: orchestrated business operations in one case and an ongoing conversational experience in the other.
Voice, documents and production engineering
My AI work at Cybernef included speech-to-text, text-to-speech and image generation using Hugging Face and OpenAI. Study Mind adds full-stack LLM work for document-based reasoning, notes and study scheduling. These are examples of connecting model capabilities to interfaces people can use.
My wider toolkit includes LangGraph, CrewAI, LlamaIndex, LangChain, RAG evaluation and observability tools. The skills section distinguishes that toolkit from the technologies attributed to each project. My practical outcomes include semantic retrieval improvements, multi-agent workflow functionality and conversational or document-based product features.
I’m Abaid Ullah, a Full Stack AI Software Engineer open to opportunities in this area.
I build LLM integrations, conversational assistants and agent workflows. My portfolio includes retrieval-augmented generation (RAG), GraphRAG, vector search, knowledge graphs and multi-agent orchestration for business applications.
Skills relevant to this role
Generative AI (Gen AI), LLMs, RAG, GraphRAG, AI agents, LangGraph, CrewAI and Model Context Protocol (MCP).
Role interests: Applied AI, generative AI applications, RAG assistants and agent orchestration.
Different AI workflows, different product needs
My AI portfolio spans document-based reasoning, conversational memory, retrieval and multi-agent automation. The AutoBot focuses on vector search and knowledge graphs; Planet AI focuses on orchestrating business tasks. Twin Companion and Study Mind demonstrate conversational and learning interfaces for AI capabilities.
Selected project experience
The AutoBot
Custom-built enterprise GraphRAG assistant application combining vector search and knowledge graphs for verifiable compliance.
My role: Full-stack + AI Systems
Planet AI
Dynamic full-stack workflow builder utilizing multi-agent AI orchestration to automate and streamline complex business tasks.
My role: Full-stack + AI Agents
Twin Companion
Conversational AI app with advanced session memory and adaptive sentiment matching that dynamically adjusts chatbot response tone.
My role: Frontend + AI Integration
Study Mind
Interactive learning platform with a NotebookLLM-style framework for document-based reasoning, notes, and study scheduling.
My role: Full-stack (Next.js + LLM)
Discuss a Generative AI & Applied AI Engineer opportunity
I have worked with international teams across the USA, Canada, Europe and Asia, including the UAE. I’m available for global teams and remote collaboration.
Send the role description, product goals and technical requirements so we can discuss the relevant experience.
My tools for this work
RAG specialist and knowledge architecture
RAG, GraphRAG, Knowledge graph RAG, Microsoft GraphRAG, LlamaIndex, Haystack, Qdrant, Pinecone, pgvector, Neo4j, LangChain, ColPali, Unstructured, Hybrid search, Reranking.
Agent architecture and orchestration
LangGraph, CrewAI, AutoGen, PydanticAI, GPT-Researcher, browser-use, OpenHands, Mem0, Model Context Protocol (MCP), Custom MCP servers, FastMCP, Text-to-SQL agents.
LLMOps, evaluation and model serving
RAGAS, Langfuse, Arize Phoenix, LangSmith, Instructor, DSPy, NeMo Guardrails, vLLM, Ollama, Structured outputs, LLM-as-judge evaluation.
Model adaptation, voice and vision
Unsloth, QLoRA, Axolotl, LiveKit, Whisper, YOLO, ComfyUI, Hugging Face, OpenAI, Vapi, ElevenLabs, OpenCV, OCR.