sawradip0@gmail.com

Hi, my name is

Sawradip Saha.

I create intelligence.

I'm a machine learning engineer specializing in building complex data processing and model training pipelines. Currently, I'm focused on contributing to open-source projects , in addition to my role at MagicMind.

About Me

I am Sawradip Saha, a machine learning engineer and researcher from Dhaka, Bangladesh. I did my BSc in Mechanical Engineering at BUET, but most of my work has been in AI, especially deep learning and agentic systems. I have coauthored several papers in IEEE conferences and journals on computer vision, activity recognition, biomedical signals, and document analysis, and I enjoy taking ideas all the way from theory to working systems.

Professionally, I have worked as a founding or early engineer at AI-focused startups, building large-scale systems for speech, vision and LLM-based workflows. I contributed to Google Summer of Code 2023 with Intel's OpenVINO Toolkit, and I actively contribute to open source projects such as OpenVINO, PyTorch and Skorch.

Right now I am building RunAgent, an infrastructure platform for deploying AI agents into real software products. I am especially interested in research at the intersection of LLM agents, tools, and systems: how to make agents reliable, controllable and efficient enough to be used in production.

Here are a few technologies I've been working with recently:

  • PyTorch
  • chatGPT
  • Cuda
  • GPT3
  • C#
  • Dall-E2
Headshot

Research Publications

Total Citations: 43
View All Publications on Google Scholar

Some of my achievements

  • Contributor at OpenVINO Toolkit, Intel

    Google Summer of Code 2023

    • Selected as a contributor for Google Summer of Code 2023 with OpenVINO Toolkit (Intel).

    • Worked on integrating Timm (PyTorch Image Models) into the Optimum-Intel library, expanding OpenVINO's support for a broader range of PyTorch models.

    • The project enhanced the bridge between Hugging Face's ecosystem and Intel's OpenVINO Toolkit, enabling developers to leverage OpenVINO's performance optimizations for Timm models.

    • Successfully contributed to the Optimum-Intel repository, making Timm models accessible through OpenVINO's optimized inference pipeline.

    • PyTorch
    • OpenVINO
    • Timm Models
    • Optimum-Intel
    • Hugging Face
    External Link
    Google Summer of Code 2023

  • Top 10

    Robi Datathon 2.0

    • The Datathon competition was organized by the country's leading digital operator Robi.

    • Datathon 2.0 was powered by 'AWS' (Amazon Web Services), 'Huawei' was the platinum sponsor, and 'Brain station' was the cloud expertise partner, reads a press release.

    • More than 2,800 participants from 11 countries took part in the Datathon 2.0 competition.

    • Following a tough competition, 100 participants took part in the final round grouped into 25 teams.

    • From those 25 teams, top 10 Teams were selected for further Presntation and Analysis.

    • Big-data
    • Time-series
    • Visualization
    • Analysis
    External Link
    Robi Datathon 2.0

  • 3rd Position

    IEEE VIP Cup 2022

    • More than 60 teams from different universities around the world participated, top three teams competed in the final round at ICIP 2022.

    • We built a generalized detector for identifying real, synthetic and partially manipulated image by GANs as well as Diffusion models.

    • Our Team Sherlock co-mentoredby me achieved 3rd Position and was awarded with a prize money of $1500.

    • Pytorch
    • GANs
    • Synthetic
    • Detection
    External Link
    IEEE VIP Cup 2022

  • 2nd Position

    Deep Learning Sprint 2022

    • 59 teams from all over Bangladesh as well as overseas participated through Kaggle.

    • We built a Wav2Vec2.0 based Bangla text‑to‑speech language model, that can correctly transcript Bangla Speech.

    • Our Team Ekush achieved the 2nd Position and was awarded with a prize money of 60000 Taka.

    • Speech-2-text
    • Wav2Vec2
    • Augmentation
    • Nemo
    External Link
    Deep Learning Sprint 2022

  • 1st Position

    IEEE VIP Cup 2021

    • More than 50 teams from different universities around the world participated.

    • We competed against Hong Kong Polytechnic University and University in Moratuwa, Sri Lanka teams in the final round.

    • We built an In-bed Human Body Key-point and Pose detection System using Long Range InfraRed(LWIR) images at adverse lighting and covered conditions.

    • Our Team Samaritan achieved 1st Position and was awarded with a prize money of $5000.

    • Augmentation
    • Semi-supervised
    • Heatmap
    • Pose
    External Link
    IEEE VIP Cup 2021

Where I've Learned

B.Sc in Mechanical Engineering

Bangladesh University of Engineering and Technology

Dhaka, Bangladesh

May 2018 - April 2023

  • The minimum credit hours requirement for this B. Sc. Engg. (Mech.) degree is 161.0.
  • All details about this undergraduate degree can be found here
  • CGPA: 3.4+

Where I've Worked

Founder @ RunAgent AI

May 2025 - Present

  • Founded and built RunAgent, an agent-native cloud platform that makes it easy for developers to run AI agents in production.
  • Architected secure, scalable micro-VM–backed services with full streaming support for Python-based agent deployment.
  • Developed language-agnostic SDKs with built-in authentication, observability, and memory capabilities.
  • Enabled seamless integration of AI agents into real applications and workflows without infrastructure complexity.

What's Next?

Get In Touch

I am passionate about continuing to grow and develop my skills in the field of machine learning and am open to discussing any potential projects, including open-source projects, or opportunities.