I'm a freelance AI developer with 5+ years of experience, specializing in the full lifecycle of LLM systems, from architecture and design through development, deployment, evaluation, and ongoing maintenance, with production observability and monitoring built in. I build reliable multi-agent systems and hybrid RAG pipelines, with hands-on expertise in open-source LLMs including GLM-5.2, Gemma 4, and gpt-oss-120b. My work also covers LLM system evaluation and traceability: defining quality metrics, assessing retrieval and generation, tracing agent and tool execution, comparing experiments, and diagnosing failures. Previous work includes AXIS, an AI meeting intelligence platform with meeting recording, transcription, task extraction, and contextual AI chat.

I can also build and ship frontend applications using AI-assisted and vibe-coding workflows, although frontend development is not my primary specialization.

I help clients design, build, deploy, and maintain end-to-end AI products, including:

• Open-source LLM selection, integration, serving, and optimization.
• Multi-agent systems with task decomposition, tool use, context management, and orchestration.
• Hybrid RAG pipelines with dense and sparse retrieval, reranking, generation, and evaluation.
• LLM evaluation and traceability for quality measurement, experiment comparison, failure analysis, and production monitoring.
• Backend services, data pipelines, secure product integration, and AI-assisted frontend delivery.

Alongside my freelance projects, I dedicate significant time to AI Talks, a volunteer-run, bilingual community focused on practical, applied AI. We organize free weekly sessions where community members share systems they have built or topics they have studied, followed by open discussion and questions.

• Help organize, host, and present sessions on production RAG, multi-agent systems, LLM costs, workflow automation, and practical AI engineering.
• Help keep the community open by publishing session material in both English and Persian, without paywalls or registration.
• Grew the initiative to 55+ sessions, 20+ speakers, 20+ topics, and 50+ bilingual session write-ups.
• Presented or contributed to 10+ sessions since joining as a speaker in Session 16.

🌐 AI Talks Hub | My sessions

Built an academic AI community with workshops, study circles, and a two-day conference.

🌐 Website | YouTube | GitHub

Built an end-to-end meeting intelligence product across a React + TypeScript web app, a Chrome MV3 recording extension, and Supabase Edge Functions. The platform supported meeting capture and upload, AssemblyAI transcription, structured task extraction, persistent contextual AI chat, and tenant-scoped meeting workflows.

Message-driven system that performs retrieval over embedded organizational data to generate LLM-based responses via broker queues.

ETL DAGs using Apache Airflow for data embedding and summarization, with caching and timestamp-based indexing to avoid redundant processing.

Scalable, fault-tolerant workflows using Temporal for async data processing, deduplication, ETL orchestration, and event-driven microservice integration.

Custom LLM fine-tuning and pipeline development for community violation detection and automated reporting.

Multi-agent LLM applications using CrewAI and LangChain for dynamic analysis and orchestrated decision-making.

Python analytics library for behavioral insights, including graph analytics and activity-based segmentation.

• Languages: Python, TypeScript, JavaScript, SQL, LaTeX
• AI/LLM: OpenAI API, open-source LLMs, hybrid RAG, multi-agent systems, llama-index, LangChain, CrewAI, LLM evaluation and traceability
• Data & Infra: Supabase, PostgreSQL, Qdrant, Neo4j, MongoDB
• Pipelines & Workflow: Apache Airflow, Temporal, RabbitMQ
• Frontend & Product: React, Vite, Tailwind CSS, Chrome Extensions (MV3)
• Tooling: Docker, Git

• ✉️ Email: dadgaramin96@gmail.com
• 💼 LinkedIn: linkedin.com/in/mramin22
• 💬 Discord: mramin22#1669
• 🐦 Twitter: @mramin22

"After everything, what remains is kindness — so don't hesitate to help others." 😊