Building production AI systems — and teaching the next generation how to build them.

I’m a Senior Data Scientist & Senior Full-Stack Software Engineer working across the entire lifecycle of intelligent systems — from research and experimentation to distributed data pipelines, cloud-native deployment, MLOps, and long-term operation of production LLM platforms.

My work spans multi-agent orchestration, hybrid retrieval, MCP-based tool ecosystems, and large-scale data engineering, backed by the discipline that makes ML safe to ship: experiment tracking, model versioning, CI/CD for ML, containerized & orchestrated serving, monitoring & drift detection, and evaluation pipelines. I treat models as software: tested, observable, reproducible, and deployed on infrastructure that scales.

Beyond engineering, I’m in love with teaching. I’ve taught at universities as an instructor, supervised students through their theses, and design active learning materials that turn complex AI and software engineering concepts into things students can actually build.

M.Sc. in Computer Science (AI), first-ranked in cohort. Published in IEEE Access and Neural Computing & Applications, with a Springer book chapter on LLMs in healthcare. Open to collaborations and senior IC roles.

Research → Prototyping → Productionization → MLOps → Monitoring → Teaching what I learned

• CardioVIT-Distill: Hardness-Aware Data Splitting and Knowledge Distillation for Robust Myocardial Abnormality Detection.  Asadolahi, M., et al.  2026.  Under review at Scientific Reports.
• Memory Reward Inflation in Self-Improving LLM Agents.  Asadolahi, M., et al.  2026.  Under review at Applied Intelligence.  DOI:10.48550/arXiv.2608.00017
• A hybrid LSTM-GRU model for stock price prediction.  IEEE Access, vol. 13 (2025), pp. 117594–117618.  ISSN 2169-3536.  DOI:10.1109/ACCESS.2025.3586558
• Explainable Large Language Models in Healthcare Applications, Springer, 2026.  Contributed chapter: Legal and Regulatory Considerations (Ch. 11, pp. 287–299).  DOI:10.1007/978-3-032-15088-2_11
• Twin Delayed Stochastic Actor-Critic.  Neural Computing and Applications, June 2023.  DOI:10.21203/rs.3.rs-3041837/v1
• CypherRAG — Teaching LLMs to Query Knowledge Graphs for Graph-Retrieval-Augmented Generation.  In preparation — method validated, manuscript drafting.

"If you can't explain it to a student, you don't understand it well enough to ship it."

• University instructor — taught Artificial Intelligence, Software Engineering, and Programming Languages at the university level.
• Thesis supervision & mentorship — guiding students through their theses and their first steps in ML research and production engineering.
• Active learning materials — creating open, hands-on course materials and repositories designed to be built with, not just read: SRB-University-LLM-Course-Spring-2026.
• Open source as teaching at scale — contributor to mem0ai, the memory layer for AI agents, with 45+ public repositories across agentic AI, RAG, RL, meta-learning, and optimization.

Claude CLI ↔ OpenAI Proxy FastAPI proxy that translates OpenAI-format requests to a local Claude CLI subprocess. Tool-call reconstruction, JSON-schema augmentation of system prompts, structured 429/504 error handling, full test suite. Lets any OpenAI-SDK tool target Claude without cloud keys.   FastAPI · async Python · tool-use · OpenAI SDK

Hybrid Agentic RAG — semantic + BM25 Hybrid retrieval combining dense embeddings, BM25 lexical search, and reranking inside an agentic loop — the pattern behind production RAG that doesn’t fall over on rare terminology.   hybrid retrieval · rerankers · LangChain

InkFlow  ·  MIT Real-time capture and analytics platform for GitHub Copilot Chat and Claude Code conversations. Watches VS Code sessions, records every prompt / response / tool call / file edit, and stores them in a searchable Postgres-backed dashboard. Built for compliance, audit, and team knowledge preservation.   TypeScript · React 19 · Express 5 · PostgreSQL 17 · Docker

MAML for Scientific Claim Detection  ·  MAML in NLP Model-Agnostic Meta-Learning applied to low-resource NLP tasks — few-shot adaptation for claim verification.

Twin Delayed Stochastic Actor-Critic (TD-SAC) Companion code to my Neural Computing & Applications (2023) paper. Stochastic off-policy actor-critic with twin-delayed critic updates.

_{Open to collaboration and always happy to talk about teaching}