Freelancer logo
Log In Sign Up
Post a Project

EXPLORE

Engineering Machine Learning (ML) Mathematics Matlab and Mathematica Mechanical Engineering
Profile cover photo
User Avatar

Sajjad T.

@sajjadtaghvaeifr

monthly-level-two.svg
4.8
196

8.2

8.2

90%

PhD in Mechanical Engineering AI/simulink

$50 USD / Hour

・

Flag of

United Arab Emirates (9:41 AM)

・

Joined on September 8, 2018

・

38 Recommendations

$50 USD / Hour

・

I have phd in mechanical engineering from department of bioengineering and robotics from Tohoku university and I lead a team of expert engineering with the following skills: ***** AI Engineering ***** Machine Learning Models ETL / ELT Pipelines MLOps Hyperparametrs Tuning Fine Tuning GenAI Prompt Engineering Hugging Face ***** Data Engineering ***** CI&CD Kubernetflow Redis Vector Database Hadoop ecosystem GCP (Dataflow, BigQuery) Containerization (Docker) ***** Mechanical Engineering***** Solidworks Ansys Mechanical design Inventor Autocad Machine design Matlab Simulink Vibration Control Stress analysis ***** Electrical Engineering ***** Signal processing Image processing Power electronic Power transmission lines Kalman filters Electronics Conrol systems Identification ***** Computer engineering ***** Programming with python, Java, cpp, etc. Neural networks deep learning convolutional neural network artificial intelligence object detection web design and development iOS and Android apps

Verifications

On time

90%

On budget

89%

Accept rate

72%

Repeat hire rate

14%

Freelancer Matlab and Mathematica Engineers United Arab Emirates
See all Matlab and Mathematica Engineers in United Arab Emirates

Portfolio

Portfolio

13951471
Start by defining the end‑to‑end pipeline: a commit or pull request in your GitHub repository triggers a workflow that packages the user prompt and any per‑second descriptions into a validated payload. The workflow validates input (length, sentence structure, required fields) and then calls the video generation API, passing authentication credentials stored as encrypted secrets. The API returns a task identifier immediately and later provides the generated asset via a callback or a polling endpoint; the workflow records the task state and any intermediate metadata in the repository or an attached database. Next, implement the runtime interactions and error handling: the workflow should send a single, well‑formed JSON payload to the model endpoint and handle synchronous responses (task accepted / rejected) and asynchronous completion notifications. On completion, download the resulting video and any auxiliary files (thumbnails, logs, JSON metadata), verify checksums, and run lightweigh

Video Generation Bot for marketing

Start by defining the end‑to‑end pipeline: a commit or pull request in your GitHub repository triggers a workflow that packages the user prompt and any per‑second descriptions into a validated payload. The workflow validates input (length, sentence structure, required fields) and then calls the video generation API, passing authentication credentials stored as encrypted secrets. The API returns a task identifier immediately and later provides the generated asset via a callback or a polling endpoint; the workflow records the task state and any intermediate metadata in the repository or an attached database. Next, implement the runtime interactions and error handling: the workflow should send a single, well‑formed JSON payload to the model endpoint and handle synchronous responses (task accepted / rejected) and asynchronous completion notifications. On completion, download the resulting video and any auxiliary files (thumbnails, logs, JSON metadata), verify checksums, and run lightweigh

Video Generation Bot for marketing

Start by defining the end‑to‑end pipeline: a commit or pull request in your GitHub repository triggers a workflow that packages the user prompt and any per‑second descriptions into a validated payload. The workflow validates input (length, sentence structure, required fields) and then calls the video generation API, passing authentication credentials stored as encrypted secrets. The API returns a task identifier immediately and later provides the generated asset via a callback or a polling endpoint; the workflow records the task state and any intermediate metadata in the repository or an attached database. Next, implement the runtime interactions and error handling: the workflow should send a single, well‑formed JSON payload to the model endpoint and handle synchronous responses (task accepted / rejected) and asynchronous completion notifications. On completion, download the resulting video and any auxiliary files (thumbnails, logs, JSON metadata), verify checksums, and run lightweigh

Video Generation Bot for marketing

13946814
This workflow automates invoice processing end‑to‑end: it triggers on file updates in Google Drive, downloads the file, and routes it through a rules‑based Switch to either an internal AI Agent path (for text‑based PDFs and human review) or an external OCR path using Mistral (upload, signed URL if required, and retrieve OCR results). Both paths normalize their outputs and merge into a single stream that is enriched and validated by an OpenAI chat model to extract structured invoice fields (order ID, date, line items, totals, taxes). The final structured data is appended to a spreadsheet, the original file is archived to a processed folder, and the incoming file can be deleted to prevent reprocessing. Key implementation notes: ensure binary handling is correct (downloaded files must appear under the exact binary key, e.g., binary.data, or use Move Binary Data to convert JSON→binary), enable Send Binary Data on HTTP upload nodes and do not manually set multipart Content‑Type headers so

AI Invoice Agent

This workflow automates invoice processing end‑to‑end: it triggers on file updates in Google Drive, downloads the file, and routes it through a rules‑based Switch to either an internal AI Agent path (for text‑based PDFs and human review) or an external OCR path using Mistral (upload, signed URL if required, and retrieve OCR results). Both paths normalize their outputs and merge into a single stream that is enriched and validated by an OpenAI chat model to extract structured invoice fields (order ID, date, line items, totals, taxes). The final structured data is appended to a spreadsheet, the original file is archived to a processed folder, and the incoming file can be deleted to prevent reprocessing. Key implementation notes: ensure binary handling is correct (downloaded files must appear under the exact binary key, e.g., binary.data, or use Move Binary Data to convert JSON→binary), enable Send Binary Data on HTTP upload nodes and do not manually set multipart Content‑Type headers so

AI Invoice Agent

13943806
The workflow is best described as a coordinated automation system in which Telegram serves as the entry point, and every incoming message—whether text, voice, or image—is routed through a switch that determines how it should be processed. Voice messages are downloaded and transcribed, while other inputs move directly to an AI agent that connects with several specialized models responsible for tasks such as cryptocurrency analysis, crude‑oil pricing, and precious‑metal data retrieval. Once the AI produces its interpretation or answer, the workflow can optionally summarize the content, convert it into speech through an external audio service, and return either text or audio back to the user on Telegram, creating a seamless end‑to‑end assistant capable of understanding, analyzing, and responding across multiple formats.

Market Analyzing agent

13926432
This workflow is a compact data‑scraping and automation pipeline built in n8n: when the workflow is executed, one branch sends a request to Firecrawl to scrape structured content from a target webpage, then passes that cleaned text to an LLM message node for summarization or analysis before sending the result through Gmail authentication as an email. At the same time, a second branch downloads an Excel file from Google Drive using Drive authentication, reads it with the Excel extractor, evaluates conditions inside the sheet, and—if the data meets your criteria—also triggers a Gmail message; otherwise it stops cleanly. Together, these two paths show how n8n can combine web scraping, AI processing, cloud‑file access, and conditional logic into a single automated workflow that reacts to both online content and spreadsheet data.This workflow is a compact data‑scraping and automation pipeline built in n8n: when the workflow is executed, one branch sends a request to Firecrawl to scrape stru

n8n data scraping

13808393
This professional Streamlit-based speech recognition application leverages the Qwen3-ASR-1.7B model to provide secure, GPU-accelerated audio transcription with an intuitive web interface. The application features a complete authentication system with username/password login, supports both MP3 and WAV audio formats, and offers customizable transcription parameters ranging from 128 to 8192 tokens to accommodate various audio lengths. Users can upload audio files through a beautiful gradient-themed interface with glassmorphism effects, preview their audio before processing, and receive comprehensive results including auto-detected language, justified text formatting, word and character counts, and processing metrics. The sidebar provides detailed information about the Qwen3-ASR model's capabilities and limitations, while offering real-time configuration options and estimated output lengths. Results can be downloaded as plain text transcripts or detailed reports with metadata, making it id

ASR & STS

This professional Streamlit-based speech recognition application leverages the Qwen3-ASR-1.7B model to provide secure, GPU-accelerated audio transcription with an intuitive web interface. The application features a complete authentication system with username/password login, supports both MP3 and WAV audio formats, and offers customizable transcription parameters ranging from 128 to 8192 tokens to accommodate various audio lengths. Users can upload audio files through a beautiful gradient-themed interface with glassmorphism effects, preview their audio before processing, and receive comprehensive results including auto-detected language, justified text formatting, word and character counts, and processing metrics. The sidebar provides detailed information about the Qwen3-ASR model's capabilities and limitations, while offering real-time configuration options and estimated output lengths. Results can be downloaded as plain text transcripts or detailed reports with metadata, making it id

ASR & STS

This professional Streamlit-based speech recognition application leverages the Qwen3-ASR-1.7B model to provide secure, GPU-accelerated audio transcription with an intuitive web interface. The application features a complete authentication system with username/password login, supports both MP3 and WAV audio formats, and offers customizable transcription parameters ranging from 128 to 8192 tokens to accommodate various audio lengths. Users can upload audio files through a beautiful gradient-themed interface with glassmorphism effects, preview their audio before processing, and receive comprehensive results including auto-detected language, justified text formatting, word and character counts, and processing metrics. The sidebar provides detailed information about the Qwen3-ASR model's capabilities and limitations, while offering real-time configuration options and estimated output lengths. Results can be downloaded as plain text transcripts or detailed reports with metadata, making it id

ASR & STS

13804068
This repository belongs to the process of detecting the emotion in variety of languages such as Arabic, English and etc. Two versions of the Small Linguistic Models (SLM) such as phi 3 and GLM 4.7 are used as he backbone for feature extraction and finally the MLP with multi gated layers were used for the final classification

Sentiment anlysis

13951471
Start by defining the end‑to‑end pipeline: a commit or pull request in your GitHub repository triggers a workflow that packages the user prompt and any per‑second descriptions into a validated payload. The workflow validates input (length, sentence structure, required fields) and then calls the video generation API, passing authentication credentials stored as encrypted secrets. The API returns a task identifier immediately and later provides the generated asset via a callback or a polling endpoint; the workflow records the task state and any intermediate metadata in the repository or an attached database. Next, implement the runtime interactions and error handling: the workflow should send a single, well‑formed JSON payload to the model endpoint and handle synchronous responses (task accepted / rejected) and asynchronous completion notifications. On completion, download the resulting video and any auxiliary files (thumbnails, logs, JSON metadata), verify checksums, and run lightweigh

Video Generation Bot for marketing

Start by defining the end‑to‑end pipeline: a commit or pull request in your GitHub repository triggers a workflow that packages the user prompt and any per‑second descriptions into a validated payload. The workflow validates input (length, sentence structure, required fields) and then calls the video generation API, passing authentication credentials stored as encrypted secrets. The API returns a task identifier immediately and later provides the generated asset via a callback or a polling endpoint; the workflow records the task state and any intermediate metadata in the repository or an attached database. Next, implement the runtime interactions and error handling: the workflow should send a single, well‑formed JSON payload to the model endpoint and handle synchronous responses (task accepted / rejected) and asynchronous completion notifications. On completion, download the resulting video and any auxiliary files (thumbnails, logs, JSON metadata), verify checksums, and run lightweigh

Video Generation Bot for marketing

Start by defining the end‑to‑end pipeline: a commit or pull request in your GitHub repository triggers a workflow that packages the user prompt and any per‑second descriptions into a validated payload. The workflow validates input (length, sentence structure, required fields) and then calls the video generation API, passing authentication credentials stored as encrypted secrets. The API returns a task identifier immediately and later provides the generated asset via a callback or a polling endpoint; the workflow records the task state and any intermediate metadata in the repository or an attached database. Next, implement the runtime interactions and error handling: the workflow should send a single, well‑formed JSON payload to the model endpoint and handle synchronous responses (task accepted / rejected) and asynchronous completion notifications. On completion, download the resulting video and any auxiliary files (thumbnails, logs, JSON metadata), verify checksums, and run lightweigh

Video Generation Bot for marketing

13946814
This workflow automates invoice processing end‑to‑end: it triggers on file updates in Google Drive, downloads the file, and routes it through a rules‑based Switch to either an internal AI Agent path (for text‑based PDFs and human review) or an external OCR path using Mistral (upload, signed URL if required, and retrieve OCR results). Both paths normalize their outputs and merge into a single stream that is enriched and validated by an OpenAI chat model to extract structured invoice fields (order ID, date, line items, totals, taxes). The final structured data is appended to a spreadsheet, the original file is archived to a processed folder, and the incoming file can be deleted to prevent reprocessing. Key implementation notes: ensure binary handling is correct (downloaded files must appear under the exact binary key, e.g., binary.data, or use Move Binary Data to convert JSON→binary), enable Send Binary Data on HTTP upload nodes and do not manually set multipart Content‑Type headers so

AI Invoice Agent

This workflow automates invoice processing end‑to‑end: it triggers on file updates in Google Drive, downloads the file, and routes it through a rules‑based Switch to either an internal AI Agent path (for text‑based PDFs and human review) or an external OCR path using Mistral (upload, signed URL if required, and retrieve OCR results). Both paths normalize their outputs and merge into a single stream that is enriched and validated by an OpenAI chat model to extract structured invoice fields (order ID, date, line items, totals, taxes). The final structured data is appended to a spreadsheet, the original file is archived to a processed folder, and the incoming file can be deleted to prevent reprocessing. Key implementation notes: ensure binary handling is correct (downloaded files must appear under the exact binary key, e.g., binary.data, or use Move Binary Data to convert JSON→binary), enable Send Binary Data on HTTP upload nodes and do not manually set multipart Content‑Type headers so

AI Invoice Agent

13943806
The workflow is best described as a coordinated automation system in which Telegram serves as the entry point, and every incoming message—whether text, voice, or image—is routed through a switch that determines how it should be processed. Voice messages are downloaded and transcribed, while other inputs move directly to an AI agent that connects with several specialized models responsible for tasks such as cryptocurrency analysis, crude‑oil pricing, and precious‑metal data retrieval. Once the AI produces its interpretation or answer, the workflow can optionally summarize the content, convert it into speech through an external audio service, and return either text or audio back to the user on Telegram, creating a seamless end‑to‑end assistant capable of understanding, analyzing, and responding across multiple formats.

Market Analyzing agent

13926432
This workflow is a compact data‑scraping and automation pipeline built in n8n: when the workflow is executed, one branch sends a request to Firecrawl to scrape structured content from a target webpage, then passes that cleaned text to an LLM message node for summarization or analysis before sending the result through Gmail authentication as an email. At the same time, a second branch downloads an Excel file from Google Drive using Drive authentication, reads it with the Excel extractor, evaluates conditions inside the sheet, and—if the data meets your criteria—also triggers a Gmail message; otherwise it stops cleanly. Together, these two paths show how n8n can combine web scraping, AI processing, cloud‑file access, and conditional logic into a single automated workflow that reacts to both online content and spreadsheet data.This workflow is a compact data‑scraping and automation pipeline built in n8n: when the workflow is executed, one branch sends a request to Firecrawl to scrape stru

n8n data scraping

13808393
This professional Streamlit-based speech recognition application leverages the Qwen3-ASR-1.7B model to provide secure, GPU-accelerated audio transcription with an intuitive web interface. The application features a complete authentication system with username/password login, supports both MP3 and WAV audio formats, and offers customizable transcription parameters ranging from 128 to 8192 tokens to accommodate various audio lengths. Users can upload audio files through a beautiful gradient-themed interface with glassmorphism effects, preview their audio before processing, and receive comprehensive results including auto-detected language, justified text formatting, word and character counts, and processing metrics. The sidebar provides detailed information about the Qwen3-ASR model's capabilities and limitations, while offering real-time configuration options and estimated output lengths. Results can be downloaded as plain text transcripts or detailed reports with metadata, making it id

ASR & STS

This professional Streamlit-based speech recognition application leverages the Qwen3-ASR-1.7B model to provide secure, GPU-accelerated audio transcription with an intuitive web interface. The application features a complete authentication system with username/password login, supports both MP3 and WAV audio formats, and offers customizable transcription parameters ranging from 128 to 8192 tokens to accommodate various audio lengths. Users can upload audio files through a beautiful gradient-themed interface with glassmorphism effects, preview their audio before processing, and receive comprehensive results including auto-detected language, justified text formatting, word and character counts, and processing metrics. The sidebar provides detailed information about the Qwen3-ASR model's capabilities and limitations, while offering real-time configuration options and estimated output lengths. Results can be downloaded as plain text transcripts or detailed reports with metadata, making it id

ASR & STS

This professional Streamlit-based speech recognition application leverages the Qwen3-ASR-1.7B model to provide secure, GPU-accelerated audio transcription with an intuitive web interface. The application features a complete authentication system with username/password login, supports both MP3 and WAV audio formats, and offers customizable transcription parameters ranging from 128 to 8192 tokens to accommodate various audio lengths. Users can upload audio files through a beautiful gradient-themed interface with glassmorphism effects, preview their audio before processing, and receive comprehensive results including auto-detected language, justified text formatting, word and character counts, and processing metrics. The sidebar provides detailed information about the Qwen3-ASR model's capabilities and limitations, while offering real-time configuration options and estimated output lengths. Results can be downloaded as plain text transcripts or detailed reports with metadata, making it id

ASR & STS

13804068
This repository belongs to the process of detecting the emotion in variety of languages such as Arabic, English and etc. Two versions of the Small Linguistic Models (SLM) such as phi 3 and GLM 4.7 are used as he backbone for feature extraction and finally the MLP with multi gated layers were used for the final classification

Sentiment anlysis

4.8 · 196 Reviews
196 Reviews
User Avatar
Flag of

Igor Filipe C.

5.0

€500 EUR

8 days ago

Padel Video & Image Annotation in Label Studio

Extremely professional. The data labelling work that was performed by Sajjad was excellent. Will definitely contact him for further work.

JSON
Computer Vision
Machine Vision / Video Analytics
Data Annotating
Image Analysis

Share

A
Flag of

Fayed A.

5.0

$400 USD

1 month ago

Writing Proposal

Sajjad was extremely professional and easy to work with. He delivered high-quality work on time within the budget. Communication was excellent throughout the project. I highly recommend him.

Engineering
Matlab and Mathematica
Mechanical Engineering
Machine Learning (ML)
Mathematics

+1 more

Share

M
Flag of

Mohammed Ali Mohammed A.

5.0

$1,800 USD

2 months ago

Proposal of second project

a very good freelancer tnx

Engineering
Matlab and Mathematica
Mechanical Engineering
Machine Learning (ML)
Mathematics

+1 more

Share

S
Flag of

Sotirios S.

4.8

$24 USD

5 months ago

Milestone for Powerpoints

Sajjad has very good knowledge and delivered the project on-time. We will have a new collaboration in the future.

Engineering
Matlab and Mathematica
Mechanical Engineering
Machine Learning (ML)
Mathematics

+1 more

Share

A
Flag of

Fayed A.

5.0

$600 USD

5 months ago

Deep Learning Segmentation

Sajjad delivered the project requirements with high quality. I am happy with his work and would be willing to work with him again in the future.

Engineering
Matlab and Mathematica
Mechanical Engineering
Machine Learning (ML)
Mathematics

+1 more

Share

Experience

CEO

Jan, 2021 - Present

•

5 years, 5 months

AvidMech

Jan, 2021 - Present

•

5 years, 5 months

AVIDMech will provide a bridge between Robotic, AI, and Mechanic.

Jan, 2021 - Present

•

5 years, 5 months

Education

Toho University

2009 - 2012

•

3 years

Ph. D.

Flag of

Japan

2009 - 2012

•

3 years

Publications

Detection of Unstable Periodic Orbits and Chaos Control in a Passive Biped Model

Springer

In this investigation, two chaos control algorithms based on linearization of Poincare´ map (OGY method) and artificial neural networks (ANNs) are utilized to control the motion of a passive biped model.

OPTIMAL MOTION PLANNING OF A PLANAR PARALLEL MANIPULATOR WITH KINEMATICALLY REDUNDANT DOF

Transactions of the Canadian Society for Mechanical Engineering

In this paper, an optimal motion planning algorithm and dynamic modeling of a planar kinematically redundant manipulator are considered.

HMM-based state classification of a user with a walking support system using visual PCA features

Taylor and Francis

In this study, we propose a method for real-time visual state classification of a user with a walking support system. The visual features are extracted using principal component analysis and classification is performed by hidden Markov models, both for real-time fall detection (one-class classification) and real-time state recognition (multi-class classification).

Freelancer logo US (International) / English
Help & Support

Freelancer

Categories Projects Contests Freelancers Enterprise Innovation Challenges AI Development Membership Preferred Freelancer Program Project Management Local Jobs Photo Anywhere Showcase API for Developers Get Verified

About

About us How it Works Security Investor Sitemap Stories News Team Awards Press Releases Careers

Terms

Privacy Policy Terms and Conditions Copyright Policy Code of Conduct Fees and Charges

Partners

Escrow.com Loadshift Warrior Forum

Apps

Apple App Store logoGoogle Play logo

Registered Users

Total Jobs Posted

Freelancer ® is a registered Trademark of Freelancer Technology Pty Limited (ACN 142 189 759)

Copyright © 2026 Freelancer Technology Pty Limited (ACN 142 189 759)