docs/Resume.txt

Arash Nicoomanesh
anicomanesh@gmail.com
Machine Learning | Generative & Agentic AI | Data Science
+98 - 9127338749
arashnicoomanesh
AI & ML Engineer with 12+ years of experience in modeling and delivering scalable, AI solutions
across healthcare, finance, and marketing. Deep expertise in traditional machine learning,
predictive modeling and Generative AI and LLM-based applications. Proven track record of
mentoring crossss-functional teams, collaborating with stakeholders, and optimizing
pipelines on both on-premises and cloud infrastructures to support high-volume deployments.
kaggle.com/arashnic
github.com/aragit
huggingface.co/Arnic
anicomanesh
Core Areas of Expertise
●​
LLM Engineering
●​ Efficient LLMs: Fine-tuning by LoRA, QLoRA, DoRA, and Unsloth. PTQuantization with GPTQ, AWQ.
●​ High-Performance Inference: TensorRT-LLM, vLLM , HF TGI, Ollama.
●​ Emerging Architectures: Sparse models and Mixture of Experts (MoE) architectures (e.g., Mixtral),
Mixture-of-Recursions and efficient SLMs such as Gemma 3 and Gemma 3n.
●​Prompt Engineering: Meta-Prompting, Adaptive, CoT, ToT, and Self-Refinement.
●​Retrieval Augmentation Generation and Vector DBs
●​ Advanced Retrieval & Reasoning Paradigms: Agentic RAG, Graph-Enhanced RAG, Hybrid Search(e.g.RRF)
●​ Context Optimization: e.g. Cross-encoder re-ranking, context compression and long-context processing.
●​ Evaluation : Evaluation pipelines using RAGAS for faithfulness, answer relevancy, and context recall .
●​ Vector Database : Pinecone, Weaviate, Faiss and Elasticsearch and advanced indexing algorithms e.g.
HNSW, IVF/IVFPQ, PQ.
●​
Autonomous AI Agents & Multi-Agent Systems
●​ Agentic System Architecture & Orchestration: Single and multi-agent systems using LangChain,
LangGraph, CrewAI, and custom orchestration logic, including scalable integration with platforms like
Vertex AI Agent Builder.
●​ Core Agentic Mechanism: e..g. Reasoning & Planning (e.g., ReAct, CoT, Tree-of-Thought), iterative planning
loops, and self-reflection mechanisms .
●​ Advanced Interoperability & Tooling: MCP, A2A, Protocol Stacking & Phased Adoption, MCP SDKs.
●​
LLMOps & Evaluation
●​ LLM-Specific Evaluation: Frameworks e.g., LM-Harness, EleutherAI's LM Eval Harness and RAGAS.
●​ MLOps & Experiment Management: LLM development workflows on GCP Vertex AI, MLflow, Weights &
Biases, and Comet ML for broader MLOps contexts.
●​ Production Deployment: CI/CD pipelines using Vertex AI, Docker and orchestration (Kubernetes).
●​ Diverse ML, deep learning modeling and time-series forecasting via pyTorch, MLlib, XGBoost, TabNet, Darts, TimesFM
and SHAP and many more, on more than 100+ real world business and industry use cases .
●​ Recommendation Engines : Applying classic ML (e.g. LightFM/DeepFM) and LLM-based cold-start solutions.
●​ Scalable Data Processing : Feature engineering and selection via RAPIDS CuDF , Polars, pySpark.
Education
●​ BSc. in Mathematics and Computer Science - Sharif University of Technology, 2001
●​ MSc. in Artificial Intelligence - Amirkabir University, Withdrawn 2005
1 of 3Professional Experience
●​ KaggleX Fellowship Program , Advisor
US, Remote, 2024 – Present
●​ Pioneered the development and deployment of a hybrid autonomous conversational AI agent, designed
for medical triage, diagnosis, support, and personalized treatment planning.
●​ Engineered LLM-driven clinical reasoning capabilities utilizing Gemini 1.5 Pro with advanced
Chain-of-Thought (CoT) prompting, complemented by Med-PaLM 2 for enhanced diagnostic accuracy and
reliability.
●​ Integrated robust medical knowledge representation through seamless entity mapping to SNOMED CT
and ICD-10 via Healthcare Natural Language AI, ensuring clinical precision and interoperability.
●​ Orchestrated complex, multi-step autonomous workflows leveraging Vertex AI Agent Builder for core
agent logic and Dialogflow CX for advanced conversational management and user interaction flows.
●​ Achieved scalable, high-performance deployment by containerizing the solution on Google Kubernetes
Engine (GKE) and optimizing stateless components with Cloud Run, ensuring robust and elastic
infrastructure for demanding healthcare applications.
●​ KeyLeadHealth , Senior Data Scientist & ML Engineer
Australia, Remote, 2020 – 2023, Feb-Mar 2024
●​ Developed a phenotyping & diagnostic plugin: Fine-tuning ClinicalBERT, BioBERT, and PubMedBERT on
de-identified EHR datasets, enabling automated extraction of key patient phenotypes and contextualized
differential-diagnosis recommendations via transformer-based embeddings.
●​ Conversational-capable drug repurposing plugin: Research initiative utilizing real-world data (RWD) and
EHRs. Implemented LLMs (Gemma2, Zephyr) with GPTQ quantization to recommend alternative
therapeutic options, facilitating hypothesis generation for off-label therapeutic use.(Azure VM Multi-GPU,
HF Transformers)
●​ Engineered time-series forecasting and classification for ICU and hospital metrics, including readmission
rates, mortality, and length of stay. Employed multivariate TS forecasting techniques using Darts
(N-BEATS,LSTM, XGB), integrating clinical variables like vital signs, lab results to enhance accuracy.
●​ Conducted research on COVID-19 diagnosis through acoustic analysis of breathing, cough, and speech
signals. Applied deep learning models to identify audio biomarkers indicative of COVID-19 infection,
achieving preliminary sensitivity and specificity on par with peer-reviewed benchmarks .
●​ KaggleX Fellowship Program , Mentor
US, Remote, 2023
Led development of Multi-turn QA chatbot to solve cold-start for product recommendation:
●​ Engages users in multi-turn dialogues to suggest products, using fine- tuned Gemma, Mistral, and Zephyr
LLMs via Hugging Face libraries.
●​ Combines Elasticsearch (keyword search) and Pinecone (semantic search) through LangChain hybrid
retrieval for relevant product/policy data.
●​ Containerized RetrievalQA service with Docker and LangServe, deployed on a Kubernetes cluster (Google
Compute Engine) for scalable inference.
●​ Cinere, AI Consultant
​
​
​
​
​
Iran, 2019
●​ Spearheaded the development and presentation of AI department transition programs, providing
strategic roadmaps and conducting collaborative brainstorming to align AI initiatives with business goals.
●​ Identified, formulated, and piloted high-impact customer analytics use cases, such as customer
segmentation, RFM analysis, CLV modeling, and churn prediction, providing actionable insights for
customer engagement and retention strategies.
●​ Designed and deployed advanced marketing analytics solutions, including multivariate, multi-step time
series forecasting models for accurate sales predictions across channels and product categories.
●​ Implemented uplift modeling (XGBoost, Darts, pylift) to optimize promotional effectiveness and
developed attribution models (MCMC, PyMC) for measuring marketing ROI.
●​ Contributed to targeting optimization (baseline approaches) and developed a proof-of-concept product
recommendation engine (LightFM), encompassing data preparation and model evaluation.
Arash Nicoomanesh Resume
2 of 3●​ Saman Bank, AI Team Lead ​
​
​
​
​​
Iran, 2018 – 2019
●​ Orchestrated the strategic planning and execution of Phase I for "Blue Bank", laying the groundwork for its
analytical framework and data-driven operations. Responsible for talent acquisition, team building, and
continuous mentorship of junior data scientists, cultivating a culture of innovation and excellence.
●​ Directed multiple high-impact predictive modeling initiatives crucial to core banking functions,
specifically optimizing payment transaction processing, mitigating loan risks, and enhancing insurance
product performance. These solutions were engineered for scalability and performance on a big data
ecosystem, leveraging PySpark, MLlib, and MMLSpark.
●​ Took the lead in conceptualizing and advocating for data modeling strategies that aimed to significantly
advance the organization's AI maturity level and foster broader AI adoption.
●​ Mellat Bank, Data Scientist & ML Engineer
Iran, 2016 – 2017
●​ Led the end-to-end development of customer predictive analytics solutions, encompassing data
ingestion, feature engineering, model training, and deployment. Specifically, engineered churn and
Customer Lifetime Value (CLTV) prediction models using PySpark (MLlib) to process and analyze large
datasets on a Hadoop distributed computing framework.
●​ Collaborated with stakeholders to define model requirements and interpret results. Additionally,
conceptualized and implemented anomaly detection algorithms for real-time transaction monitoring,
contributing to enhanced security protocols and efficient resource management.
●​ MIRAS Technologies, Data Scientist
Iran, 2015 – 2016
●​ Managed the full lifecycle of text analytics projects for Samsung, from data acquisition to insight
generation. Applied advanced Natural Language Processing (NLP) techniques, including sentiment
analysis and entity extraction, to transform raw news data into actionable intelligence for news
recommendation systems. This work provided Samsung with a deeper understanding of public
perception and content relevance.
●​ Played a key role in the creation and optimization of pretrained Persian NLP models (ParsBERT),
demonstrating expertise in deep learning for linguistic applications.
●​ Engineered and maintained a high-performance web crawling framework leveraging Scrapy, ensuring
reliable and comprehensive data collection for analytical pipelines.
●​ Fanap, Data Analyst
Iran, 2014 – 2015
●​ Utilized advanced analytical techniques to extract valuable insights from sensor data, directly contributing
to the optimization of manufacturing workflows and supply chain logistics. This analysis informed
strategic decisions, leading to demonstrable improvements in efficiency and cost reduction.
●​ Architected and implemented a comprehensive Key Performance Indicator (KPI) system, leveraging the
principles of the Balanced Scorecard to monitor organizational performance. Effectively integrated this
KPI structure into the broader CRISP-DM process, ensuring a data-driven approach to problem-solving
and project execution.
●​ Spearheaded the design and modeling of enterprise data warehouses using the industry-standard
Kimball dimensional modeling methodology, enabling robust data storage, retrieval, and analytical
capabilities for diverse business units.
Arash Nicoomanesh Resume
3 of 3