Professional Experience

AI-Ark

Dec 2024 to Oct 2025

Role: Lead Machine Learning Engineer

Leading enterprise AI solutions development at app.ai-ark.com

Core Infrastructure:

• Architected hybrid vector search system with Nomic-Embed-1.5B and Vespa/Elasticsearch
• Optimized ANN/KNN ranking profiles improving F1-score from 0.76 to 0.92 on benchmarks
• Designed Oracle DB integration pipeline processing 5M+ documents daily with BGE-m3

Model Development:

• Fine-tuned T5, BART, and Qwen models on information extraction tasks
• Implemented multi-label classifications with BERT, reaching 92% and 86% F1-score across 15 and 9 categories
• Implemented 4-bit and 8-bit quantization using GGUF and AWQ techniques, reducing model size by 75% while maintaining +90% of performance

Data Engineering:

• Automated domain-specific data collection pipeline using ZenRows and AWS Bedrock
• Reduced data preparation time by 65%

Multi-Turn Multi-User RAG Chatbot with Persistent Memory for Customer Support

• Architected conversational memory system using Qdrant vector database with thread-scoped session management and buffer memory for context-aware multi-turn interactions
• Implemented session persistence layer with Redis for storing conversation history, user interaction metadata, and maintaining state across distributed instances
• Deployed hybrid search infrastructure in Qdrant combining dense vector embeddings, sparse vectors (BM25), and TF-IDF indexing for knowledge base and FAQ retrieval
• Integrated HuggingFace Text Embeddings Inference (TEI) for production-grade embedding model deployment with optimized inference and dynamic batching

Financial Behavioral Analysis Systems | GitHub

• Led team of 6 engineers in developing classification systems for financial behavior analysis
• Designed classifier architecture achieving 95.2% pattern detection accuracy
• Built dual-classifier system (LLM & ensembles), achieving 95.2% pattern detection accuracy
• Developed nearest-neighbor engine using BAAI/bge-m3 embeddings, attaining 100% precision in financial classification
• Optimized voting classifier integrating XGBoost, RandomForest, and SVM with enhanced hyperparameter tuning
• Created production-grade feature pipeline handling complex financial ratios and missing data

Fanaavaran Farayand Farda

Jun 2024 to Dec 2024

Role: Lead AI Engineer | https://fffarda.ir

Retail Analytics - Ofogh360 | Website

• Engineered custom dataset generation pipeline with selection criteria, guided by distribution-aware EDA and decision boundary analysis
• Implemented transfer learning from YOLO-SKU110K for pseudo-labeling
• Developed data synthesis algorithm resolving class imbalance using generative augmentation and minority class oversampling techniques
• Applied few-shot learning methodology to train detection/segmentation model on 53 classes using only 204 annotated images, achieving 81% mAP
• Created custom data augmentation pipeline, reducing labeled data requirements by 40%
• Integrated transfer learning, few-shot techniques, and synthesized data into end-to-end training workflow, improving product recognition accuracy by 32% across 1,000+ SKUs
• Reduced model latency by 70% through INT8 quantization and TensorRT conversion, enabling real-time inference (50ms)
• Deployed models via TorchServe with custom handlers, supporting 200+ concurrent requests

Page Streaming Segmentation & Document Intelligence Platform

• Fine-tuned multi-lingual vision-language models for invoices OCR
• Built training pipeline with 5K+ annotated documents using Roboflow and Hugging Face datasets
• Developed Streamlit UI for real-time parsing, reduced processing time from 2 mins to 30 secs

Fasta Robotics

Nov 2023 to Oct 2025

Role: AI Researcher (Part-time)

Autonomous Warehouse Robotic Systems

• Implemented multi-robot coordination framework using ORB-SLAM3 visual-inertial mapping system on ROS2 using Intel RealSense D435i and ZED 2i cameras
• Developed VIO-LiDAR sensor fusion pipeline achieving 15cm localization accuracy in dynamic warehouse environments using Jetson Orin NX compute platform
• Deployed YOLOv12 object detection achieving 95% pedestrian detection accuracy for real-time collision avoidance
• Designed end-to-end learning framework for inertial odometry estimation, improving robustness against sensor degradation by 30% through adaptive IMU sampling rate optimization
• Integrated heterogeneous sensor streams (visual, inertial, LiDAR) for collaborative mapping and navigation decision-making across robot teams
• Implemented sensor abstraction layer for Visual-Inertial and LiDAR streams (KITTI format)
• Engineered modular VIO-LiDAR fusion framework with real-time point cloud registration
• Developed extensible SLAM backend with automated performance benchmarking

Rajaei Cardiovascular Center

May 2022 to Jun 2024

Role: Machine Learning Engineer

ResearchPulse: Literature Intelligence

• Built search architecture achieving 92% precision across PubMed/Scopus databases
• Created visualization tools for trends analysis
• Deployed a contextual QA, evidence-based chatbot reducing literature review time
• Developed a Scopus/PubMed search system with trend visualization and QA chatbot
• Managed team of 5 ML engineers in developing automated classification pipelines

Clinical Decision Framework

• Built RAG pipeline processing 200+ cases with 87% diagnostic accuracy
• Developed NLP components standardizing clinical narratives into outputs
• Integrated clinical guidelines into inference with traceable reasoning

Medical Coding Platform

• Designed validation system reducing ICD-10 coding efforts by 75% at 94% accuracy
• Built validation against guidelines, eliminating critical misclassifications

LLM Research

• Implemented hybrid retrieval system combining BM25, FAISS, and cross-encoder reranking
• Implemented semantic chunking with sliding window approach, enhancing contextual preservation for long-document understanding
• Tackled hallucinations issue with multi-routing and prompt engineering
• Fine-tuned Nemotron-3 and AYA-101 models on Casual Language Modeling and Seq-2-Seq
• Utilized instruction fine-tuning on LLAMA 3 to improve instruction following

Conversational RAG Chatbot

• Conducted feasibility study evaluating 15 LLMs (7B-70B parameters) on local deployment using Ollama, benchmarking response quality and latency
• Implemented hierarchical document chunking strategy for 150K+ pages across 11K topics, optimizing for context preservation
• Created hybrid vector store using FAISS (HNSW index) with BGE-M3 embeddings and BM25 sparse vectors
• Developed multi-route orchestration system with dynamic LLM selection based on query complexity, token budget, and specialized knowledge domains
• Engineered custom prompt templates with few-shot examples and structured output formatting for consistent response generation
• Implemented query decomposition and Fusion RAG for handling vague and edge cases to improve retrieval accuracy
• Integrated cross-encoder reranking to improve relevance scoring and improve final generation accuracy
• Built API gateway interfacing with OpenAI, Anthropic, Google (Gemini), and local Ollama endpoints with automatic failover mechanisms
• Implemented Flask-based backend with streaming responses and rate limiting
• Added user feedback collection mechanism to continuously fine-tune retrieval parameters and improve answer quality through supervised learning

BiodataBank Clinical Platform - NCIBB.ir

• Led cross-functional team of medical professionals and engineers, launching platform 2 weeks ahead of schedule and 15% under budget
• Architected RAG system for 100K+ clinical documents information retrieval system
• Optimized vector search latency by 70%, reducing average query time from 2.1s to 0.6s
• Developed differential diagnosis engine on validated medical case studies across 20+ specialties
• Created real-time research visualization system analyzing PubMed papers with QA capabilities
• Built multi-route LLM-based automated ICD-10 coding system with 91% accuracy, reducing manual coding time by 75%
• Created LLM-based explainable AI module highlighting evidence supporting diagnostic suggestions

Medical Case Generator

• Created generative AI system for clinical scenario simulation with parameter controls
• Integrated feedback mechanisms for medical education research

Automated ICD-10 Labeler

• Developed hybrid retrieval system combining sparse (BM25) and dense (FAISS) embeddings
• Implemented NLP pipeline for medical entity extraction and ontology mapping
• Engineered sliding window chunking algorithm with cross-encoder reranking
• Created validation framework interfacing with ICD-10 taxonomies

Farzan Research Institute

Oct 2021 to Jun 2024

Role: AI Research Scientist

Evidence-based RAG Pipeline for Scientific AI-Assistant

• Fine-tuned Nemotron, AYA-101 and BERT models using LoRA for mental health analysis
• Created comprehensive LLM evaluation framework with domain-specific metrics
• Optimized training on A100/H100 GPUs using DeepSpeed ZeRO-3
• Engineered hybrid retrieval (BM25, FAISS, cross-encoders) and semantic chunking
• Devised Chain-of-Thought, ReAct reasoning prompt templates
• Fine-tuned Nemotron-3 8B & AYA-101 (QLoRA, 4-bit quantization) on A100 clusters
• Implemented gradient accumulation and mixed precision training on distributed A100 clusters

AI Engineer Projects

• Built conversational QA system using RAG for CRM, reducing response time 45%
• Fine-tuned BERT/LLAMA using QLoRA improving accuracy 30%
• Built RAG-based QA chatbot using OpenAI API
• Automated FAQs and integrated shopping cart tracking for enhanced user experience

Computer Vision Engineer & Freelancer

Apr 2020 to Oct 2021

Role: Independent Contractor

• Implemented LSTM-based image captioning achieving BLEU score of 0.32
• Built CNN-LSTM handwriting recognition with 5.2% CER and 8.1% WER on IAM database
• Developed classification models using EfficientNet with 94% accuracy on Breast Cancer Wisconsin dataset
• Created Mask R-CNN instance segmentation achieving 43.5 mAP on COCO dataset | Kaggle Tutorial
• Developed PyTorch training pipeline with distributed computing and custom medical image augmentation techniques
• Built real-time ROS perception system for autonomous navigation with OpenCV | GitHub
• Published computer vision tutorials and documentation | Example
• Developed PyTorch training pipelines with distributed computing, medical image augmentation, TensorRT optimization | PyTorch Tutorial
• Built ROS2 perception systems (depth estimation, object tracking) with KITTI dataset | KITTI Tutorial | ROS & OpenCV
• Created advanced Python/CV tutorials | Python Tutorial | Segmentation Example 1 | Segmentation Example 2

Research Experience

Students’ Scientific Research Center

Jul 2024 to Sep 2024

Role: Physics-Informed AI Researcher

Transformer-Enhanced PINN for Attitude Estimation

• Built TE-PINN: transformer-physics network for IMU orientation estimation
• Designed multi-head attention for IMU data, reduced error 36.8%
• Implemented RK4 quaternion integration with uncertainty quantification
• Developed PINNs with multi-head attention, reducing attitude estimation error by 36.8%
• Embedded quaternion kinematics, rigid body dynamics, and multi-head attention
• Designed a physics-based loss enforcing rotational dynamics with angular velocity and forces
• Achieved 36.8% error reduction and robustness in high-noise, dynamic conditions
• Research selected for ICRA 2025; code and datasets publicly available | GitHub | arXiv

MIT IAIFI AI+Physics Program

May 2024 to Jun 2024

Role: Research Participant

Simulation-Based Inference for Medical Analysis

• Developed Vision Transformer-based tumor severity simulator for histopathological images
• Applied Sequential Neural Posterior Estimation (SNPE) for posterior distribution analysis
• Created noise-injection model for tumor severity levels with likelihood-free inference methods
• Developed Vision Transformer-based analysis system achieving 87% diagnostic accuracy on medical imaging datasets
• Implemented Sequential Neural Posterior Estimation reducing uncertainty bounds
• Created robust uncertainty quantification framework improving confidence calibration
• Presented research findings at cross-institutional workshop with 75+ participants from MIT, Harvard, and other institutions

Oxford Machine Learning Summer School

May 2023 to Aug 2023

Role: Competition Winner

Vision-based Cancer Detection

• Implemented transfer learning pipeline achieving 82% accuracy, ranking 1st among 120+ participants in international competition
• Developed ensemble approach combining EfficientNet and Vision Transformer models
• Published methodology and results on GitHub and Kaggle | GitHub
• Applied transfer learning to medical imaging, achieving 82% accuracy
• Ranked 1st in The Health and Medicine OxML competition track | Leaderboard | Slides

University of Tehran

Sep 2019 to Sep 2022

Role: Graduate Researcher

Deep Learning for Inertial Navigation Systems

• Developed learning-based models for real-time inertial attitude estimation | GitHub | Publication
• Developed end-to-end learning models reducing attitude estimation error by 40% across 7 publicly available datasets
• 40% improvement over traditional methods on 7 datasets (100+ km IMU measurements)
• Published results in Measurements journal (Elsevier) | IF 5.2
• Open-sourced implementation with documentation and examples [+33 Stars] | GitHub

Teaching & Academic Service

Tehran University of Medical Sciences

Sep 2022 to Present

Role: Co-Instructor

Course: Application of Technology in Research

• Designing and taught graduate-level courses in advanced search techniques, providing rigorous assessment and guidance to cultivate deep subject matter expertise
• Conducting office hours, fostering academic excellence and professional growth

Students’ Scientific Research Center

Sep 2022 to Present

Role: Instructor

Graduate and Undergraduate (B.Sc., M.D., M.Sc. and Ph.D.): +100 Students

• Developing and delivered courses emphasizing practical applications of AI and Programming
• Offering personalized guidance to students, cultivating the next generation of independent researchers

Students’ Scientific Research Center

May 2023 to Present

Role: Supervisor

• Guiding 10+ students in creating AI medical imaging tools for early and accurate disease detection, enhancing patient outcomes, and cutting healthcare costs
• Leading team in six systematic reviews on AI-powered Medical Imaging Analysis, uncovering critical insights for cancer detection and time-series forecasting advancements, fostering life-saving interventions

Students’ Scientific Research Center

Apr 2019 to Present

Role: Referee of Research Council

• Shaped impactful medical science and healthcare solutions through analyzing and assessing research proposals

University of Tehran

Sep 2021 to Sep 2022

Role: Teaching Assistant

Course: Fuzzy Logic (Graduate Level) Instructor: Dr. M.H. Sabour

• Designed and supervised projects, enhancing programming skills for 15+ graduate students
• Enhanced academic and professional growth through focused office hours, optimizing graduate students’ development with strategic support

Aviation Industry Training Center

Sep 2019 to Sep 2021

Role: Thesis Supervisor

Supervised 5 Undergraduate Theses:

• Design and Implementation of a 3 Axis CNC Machine (Spring 2021 - Fall 2021)
• Design and Implementation of Pulse Circuits Training Board (Fall 2020 - Fall 2021)
• Design, Simulation, and Building of an Aircraft Fire Extinguishing System (Spring 2020 - Fall 2020)
• Design and Implementation of Retractable Landing Gear (Fall 2019 - Spring 2020)
• Design and Implementation of a CNC Hot Wire (Fall 2019 - Spring 2020)

Aviation Industry Training Center

Sep 2018 to Sep 2021

Role: Instructor

• Instructed 11 courses on electronics, navigation, and aviation for 150+ undergraduate students
• Delivering high-caliber education, nurturing a pipeline of highly skilled aerospace professionals

Review Activity

Role: Journal & Conference Reviewer (100+ Papers Reviewed)

Full list available at ORCID and Web of Science

Outstanding Recognition

• Outstanding Reviewer, IEEE Transactions on Instrumentation & Measurement, 2023

Journals:

• IEEE Robotics and Automation Letters
• IEEE Transactions on Instrumentation & Measurement, 43 Papers
• IEEE Sensors
• IEEE Instrumentation & Measurement Magazine, 1 Paper
• IEEE Open Access Journal on Circuits and Systems, 1 Paper
• Wiley Journal of Field Robotics
• Elsevier Automatica
• Elsevier Aerospace Science and Technology, 15 Papers
• Elsevier Measurement, 6 Papers
• Springer Visual Computing for Industry, Biomedicine, and Art
• Space: Science & Technology, 4 Papers
• The Aeronautical Journal, 3 Papers

Conferences:

• International Conference on Robotics and Automation (ICRA) 2024
• International Conference on Learning Representations (ICLR) 2024
• International Federation of Automatic Control (IFAC) World Congress 2023, 1 Paper
• American Control Conference (ACC) 2023, 2024

Leadership Experience

Space Generation Advisory Council

Nov 2020 to Present

Role: Mentor for AI and Space Technology

Mentorship & Guidance:

• Provided personalized mentorship to 10+ professionals in computer vision and satellite data processing
• Offered targeted guidance and sustained support to mentees, fostering future leaders in space technology
• Conducted one-on-one sessions covering career development, research methodology, and technical skills

Curriculum Development:

• Designed comprehensive curriculum covering:
  • Deep learning fundamentals and advanced architectures
  • Computer vision for satellite imagery analysis
  • AI applications in space exploration and robotics
  • Practical implementation with PyTorch and TensorFlow
• Created structured learning pathways for professionals at different skill levels

Workshops & Training:

• Led multiple workshops on AI applications in space technology
• Topics covered: satellite image segmentation, object detection in orbital imagery, autonomous spacecraft navigation
• Created extensive training materials including:
  • Tutorial notebooks for satellite image analysis
  • Documentation on implementing CV algorithms for space data
  • Best practices for deploying ML models in space applications

Community Impact:

• Contributing to capacity building in the space technology sector
• Fostering interdisciplinary collaboration between AI and space engineering communities

World Astronomy Week (Iran)

Jan 2017 to Jan 2023

Role: Executive Member

Astronomy Outreach, University of Tehran

Jan 2021 to Sep 2023

Role: Executive Member

Iran Martial Arts Federation

Mar 2016 to Present

Role: Martial Arts Instructor

• Instructed 400+ from diverse backgrounds, cultivating essential communication and life skills while fostering physical and mental well-being
• Nurtured discipline, leadership, and teamwork within the context of martial arts for comprehensive personal development
• Black Belt Dan II

Honors, Prizes, Awards, and Fellowships

AI + Physics

IAIFI + MIT - Jul 2024

MLx Representation Learning & Generative AI

Oxford Machine Learning Summer School - Jun 2024

Outstanding Reviewer

IEEE Transactions on Instrumentation & Measurement - 2023

Recognized for exceptional contributions to the peer review process

AWS AI & ML Scholarship 2023

$4000 Valued

Endowed by: Amazon and Udacity

Criteria: Completing ML course, implementing Reinforcement Learning for optimal agent decision-making and achieving a sub-one-minute lap in AWS DeepRacer Student League. Awarded to top 1% of applicants.

Ranked 1st in OxML Competition Track

2023

Criteria: Attained the highest accuracy (82.3%) in cancer cell detection among 120+ international participants

Top 10% in M.Sc. Aerospace Engineering

2019

Criteria: Ranked in the top 10% among 5000+ participants in the National University Entrance Exam

Ranked 1st in Class 2019

University of Tehran, College of Interdisciplinary Science and Technology - 2021

Criteria: Attained the highest GPA (4.0 out of 4.0) in the class

Medalist, Iran Martial Arts Federation, National Competitions

• Gold Medalist: 2011, 2012, 2018, and 2019
• Silver Medalist: 2015
• Bronze Medalist: 2016 and 2019

Black Belt Dan II

Nearu Martial Arts, 2015

Achieved due to advanced mastery in martial arts techniques and philosophy, requiring a decade of dedicated training, highlights exceptional skill, knowledge, and commitment