You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Which PyTorch Foundation projects is the nominee familiar with?
PyTorch
vLLM
DeepSpeed
Ray
Helion
Safetensors
Other
If 'Other' was selected, please specify the project name
No response
How has the nominee contributed to the community?
Open source contributions
Technical content creation
Community leadership
Events or workshops
Research or academic work
Mentorship
University or student outreach
Translation or localization efforts
Other
If 'Other' was selected, please specify
Policy and standards work on AI safety/security
Please describe the nominee’s contributions to PyTorch Foundation projects and communities
My engagement with the PyTorch Foundation spans open-source contributions, original tooling, technical writing, community support on official platforms, conference speaking, and policy/standards advocacy. The links below back every claim.
--Open Source Contributions to PyTorch Foundation Projects
torchdiag — PyTorch model health diagnostics toolkit (creator and maintainer)
A lightweight, pip-installable diagnostic library that brings SRE-style observability to PyTorch model training: gradient health checks, dead-neuron detection, training-step verification, and memory profiling. Five commands, zero configuration, MIT-licensed.
CI: Tested across Python 3.9 / 3.10 / 3.11 / 3.12 via GitHub Actions
Installable: pip install torchdiag
pytorch/tutorials — Open Pull Request #3908
Authored "Visualizing the Mathematics Behind Neural Network Training," a beginner tutorial that traces how linear algebra, calculus, and optimization map directly to torch.nn.Linear, torch.autograd, and torch.optim. Every concept is verified with runnable code (torch.allclose) so learners see the math, not just read about it. CI green, awaiting review.
pytorch/pytorch — Inductor compiler diagnostics
Submitted precise root-cause analyses on two open Inductor bugs, including a verified end-to-end reproduction and a tested fix with documented tradeoffs:
Newsletter — Technikal (technikal.substack.com) — PyTorch series:
"Why PyTorch Isn't a Framework Anymore" — the Foundation's ecosystem consolidation story
"What Actually Happens When You Call loss.backward()" — chain rule and autograd internals
"The Compiler Wars Inside PyTorch" — torch.compile vs TorchScript vs ONNX vs TensorRT vs vLLM
"The Most Important Piece of PyTorch Engineering Nobody Talks About" — deep dive on the dispatcher
Medium technical articles: https://medium.com/@aditya_mehra
"5 PyTorch Commands That Show You What Your Model Is Actually Doing" — practical model inspection
"Your PyTorch Model Is Using 3x More GPU Memory Than You Think" — memory profiling guide
Dev.to: https://dev.to/aditya_mehra
"I Built a Diagnostic Toolkit for PyTorch Because I Was Tired of Guessing Why Models Fail" — torchdiag deep-dive
Grace Hopper Celebration 2025 Talk Repository: https://github.com/AddyM/Math_behind_ML
"The Math Behind the Magic: Understanding the Role of Mathematics in Deep Learning" — slides, speaker materials, references explicitly mapped to PyTorch APIs
IEEE-USA InSight article on AI legislation in Congress (published, advocacy work bridging PyTorch ecosystem reality and federal AI policy) https://ieeeusa.org/committees/aipc/
IEEE-USA AI Policy Committee — Co-Chair Roles & Contributions:
Co-Chair, Agentic AI Subcommittee — leading position statement
development on AI agent safety, accountability, and security
Co-Chair, Mental Health AI Working Group — drafting policy guidance
on responsible deployment of AI in mental health contexts
NIST CAISI Submission Co-Author — contributed to IEEE-USA's formal
response on agentic AI security to the NIST Center for AI
Standards and Innovation
Community Engagement on Official PyTorch Platforms
https://discuss.pytorch.org/
my profile: https://discuss.pytorch.org/u/aditya_mehra/summary
PyTorch Discuss Forum — multiple substantive technical answers, trust level promoted by the platform based on quality of contributions:
Multi-node CPU DDP with SLURM (distributed training)
MPS / torch.compile performance on Apple Silicon
libtorch 2.10 vs 2.1 CPU performance regression diagnostics
RTX 5060 Ti install troubleshooting
Conv2d batch-vs-channel shape confusion
libcudnn / libnccl link errors with prebuilt libtorch on Linux
LinkedIn Community — "PyTorch Fundamentals — Learn, Build, Contribute" Group (Founder) https://www.linkedin.com/groups/19024016
A free learning community for PyTorch practitioners, supporting a planned weekly workshop series focused on accessibility for learners without traditional math backgrounds.
vLLM SIG CI meeting — attended; engaged on continuous integration infrastructure for the vLLM project (a PyTorch Foundation project).
Conference Speaking
Grace Hopper Celebration 2025 — "The Math Behind the Magic: Understanding the Role of Mathematics in Deep Learning" (Chicago)
PyCon US — Python community
PyCon India — Python/ML community
PyCon NZ (Kiwi PyCon) — Python community
PyCon Poland
PyCon Africa
PyCon Spain
PyOhio
ISCC (International Symposium on Cluster Computing)
Mentorship & Educational Outreach
Eagle Online Academy — Volunteer English teacher supporting Afghan women and refugees in the USA, an initiative I am exploring how to expand with accessible PyTorch fundamentals materials for first-generation ML learners.
NASA Conrad Challenge — Student mentor (high-school innovation program)
Devpost / hackathon judging — Multiple hackathons across innovation and AI tracks
ACM Peer Reviewer Certified — Reviewer for ACM Transactions on AI Security and Privacy (TAISAP)
Research, Standards & Policy Advocacy
IEEE-USA AI Policy Committee — Co-Chair of the Agentic AI subcommittee and the Mental Health AI position statement working group
IEEE P7014 Standard — Contributor
NIST CAISI Submission on agentic AI security — co-author
DOE RFI Response on AI — contributor
6 peer-reviewed publications in distributed systems and ML
2 UK Design Patents
Python Software Foundation Member
IEEE Senior Member
Distinctive contribution profile: I bring an SRE/production-infrastructure perspective to the PyTorch ecosystem — building tools that automate the observability practices ML teams rarely apply, contributing technical diagnoses on the compiler stack, and creating accessible educational content that connects mathematical foundations to PyTorch's actual APIs.
Why does the nominee want to become a PyTorch Foundation Ambassador?
In 18 years of building distributed systems, I have watched technical communities succeed or fail based on a single attribute: how accessible they make their internals to people who do not arrive with the field's prerequisites. PyTorch is the rare community that built its framework around making the underlying mathematics observable — every weight printable, every gradient inspectable, every operation in the computational graph traceable. That design choice is an invitation. The community has not fully cashed it in yet, and I want to help.
My motivation rests on three threads that have converged over the past year:
Translation. When I presented "The Math Behind the Magic" at Grace Hopper Celebration 2025, the room that responded most strongly was not the experienced ML engineers — it was the career switchers, the working professionals from adjacent fields, and the students who had bounced off PyTorch tutorials that assumed three semesters of calculus. PyTorch's eager execution model makes deep learning mathematics teachable in a way no other framework does. I want to build the educational materials, workshops, and content that actually use this property — turning the framework itself into the teacher.
Production reliability. Most ML failures I have debugged in production are not novel ML problems. They are old distributed systems problems with new names: silent failures, missing observability, untested error paths. I built torchdiag because the SRE practices that took the web industry a decade to internalize have not yet reached ML teams. As an ambassador, I want to bring that perspective into the PyTorch community more systematically — through tooling, writing, and direct engagement with practitioners building real systems.
Policy bridge. As Co-Chair of the IEEE-USA AI Policy Committee subcommittees on Agentic AI and Mental Health AI, I sit at conversations where federal policy gets made about AI systems — often by people who have never written import torch. The PyTorch Foundation deserves an ambassador who can carry technical reality into those rooms and bring policy reality back into the engineering community.
An ambassador role is not a destination — it is the formal community standing that lets the writing, teaching, tooling, and advocacy work I am already doing reach further. I am committed to this work regardless of the outcome of this application. Being an ambassador would let me do it with the visibility, network, and resources to make it stick.
How would this nominee contribute as a PyTorch Foundation Ambassador?
My plan as an ambassador is to expand work I am already doing across six areas, not to start something new for the sake of the role. Each of the commitments below builds on something already in motion.
Content creation. I write a technical newsletter at technikal.substack.com with a running series on PyTorch internals — the dispatcher, autograd, the compiler stack, and the Foundation's ecosystem consolidation. I will continue this series at a steady cadence, with deeper pieces focused on production reliability, model debugging, and the parts of the framework most practitioners never look at. I will also continue publishing on Medium and Dev.to to reach developers who do not read newsletters.
Educational outreach. I founded the "PyTorch Fundamentals — Learn, Build, Contribute" LinkedIn community as a learning space for practitioners who do not come from a traditional math or research background. As an ambassador I plan to run a virtual workshop series through this group, drawing on the material from my Grace Hopper Celebration 2025 talk that mapped the mathematics of deep learning directly to PyTorch APIs. The intent is making the framework accessible to learners the typical PyTorch tutorial assumes too much from.
Organizing community events. I recently relocated to the San Francisco Bay Area, which puts me close to a concentration of PyTorch contributors, researchers, and production engineers. I would like to help organize in-person PyTorch meetups in the Bay Area — both general-purpose and focused on niches that are underrepresented at existing events, such as ML in production, model observability, and the SRE / reliability lens applied to training systems.
Mentoring contributors. I just opened my own first PR to pytorch/tutorials (#3908) and went through the workflow of forking, branching, contributing, and engaging with maintainers. I would like to help others through that same process — guiding people in my LinkedIn community and through Eagle Online Academy's network from "I have never opened a PR" to "I have an open PR on a PyTorch Foundation project." Outside the PyTorch ecosystem I have already mentored students through the NASA Conrad Challenge and judged hackathons, so this is a continuation, not a new commitment.
Technical advocacy. I built torchdiag, a small open-source diagnostic toolkit for PyTorch models (pip install torchdiag), bringing SRE-style observability practices to model training. I plan to keep developing it — adding diagnostics suggested by the community, integrating it with common training workflows, and using it as a vehicle for teaching people what to look for when their models do not learn. I will also continue submitting diagnostic comments and fixes on open PyTorch issues, particularly in the Inductor and torch.compile area where I have already started engaging.
Supporting local and global AI communities. As Co-Chair of the IEEE-USA AI Policy Committee subcommittees on Agentic AI and Mental Health AI, I am regularly in policy conversations where decisions about AI systems get made by people who have never written PyTorch code. I would like to carry the PyTorch community's technical reality into those rooms — through NIST, DOE, and IEEE submissions — and bring back what I learn into the engineering community. I have also spoken at PyCon events in multiple countries and will continue to engage with the global Python and ML community through that circuit.
The thread connecting all of this is a single commitment: making PyTorch accessible to people who would not otherwise find a way in — whether they lack the math background, the production experience, the policy context, or the geographic proximity to the existing centers of activity. An ambassador role would give that work a formal platform and a community to do it with.
Ambassador Focus Areas
Community Building
Technical Education
Open Source Contributions
Research & Academia
Developer Advocacy
Mentorship
Events & Workshops
Content Creation
Regional Community Growth
Other
If 'Other' was selected, please specify
No response
Primary Community Region
North America
If 'Other' was selected, please specify
No response
Primary Country During Ambassador Term
USA
Additional Information (Optional)
A bit of professional and supplementary context that did not fit cleanly in the contributions section but supports the case above.
Production ML background. The SRE and production-reliability lens I bring to PyTorch is grounded in actual work, not theory. I spent eleven years at VMware on ESXi / VMkernel and cloud automation, then went on to senior engineering roles at American Express, Fungible (acquired by Microsoft), Bank of America (Senior Engineering Manager, ML — leading a team of 15+ on fraud detection and retrieval-augmented systems), and Fannie Mae (Senior Software Architect, Enterprise Innovation — designing RAG and agentic AI pipelines with LangChain and LangGraph). Most of the PyTorch-in-production patterns I write about come from systems I or my teams have shipped.
Distributed systems and infrastructure foundation. Before ML became my primary focus, I worked on hypervisor and cloud infrastructure problems for over a decade. This is why the SRE lens I bring to model training feels natural — gradient health checks, dead-neuron detection, and training-step verification are the same observability practices used for monitoring service health, applied to a different layer.
Earlier open-source and community work.
Contributed to Optuna, Haystack, sklearn-pandas
Author of airegex — a small CLI tool in Go
Built BridgeAI — a WhatsApp-native AI tutor concept for refugees
Contributor to the awesome-ai-explainability curated resource list
Co-founder of Infinict (Web3/DeFi, acting CTO)
Speaking history.
PyCon US, PyCon India, PyCon NZ (Kiwi PyCon), PyCon Poland, PyCon Africa, PyCon Spain
PyOhio (regional Python conference)
Grace Hopper Celebration 2025 — The Math Behind the Magic
ISCC (International Symposium on Cluster Computing)
DC Startup & Tech Week panel on Web3 and AI
CPI Global B2B Payments Summit
Recognition and credentials not already listed above.
IEEE Senior Member (a grade requiring ten or more years of professional engineering and significant contributions to the field)
Python Software Foundation Member
ACM Peer Reviewer Certified (Transactions on AI Security and Privacy)
Two UK design patents
Approximately six peer-reviewed publications
Application Type
Self-nomination
Your Name
Aditya Mehra
Your Email
aditya.mehra@ieee.org
Please confirm the following:
Nominee Full Name
Aditya Mehra
Affiliation / Organization
IEEE , USA
Nominee Email
aditya.mehra@ieee.org
Current Location
Cupertino, USA
Years of Community Involvement
5+ years
Which platforms or profiles would you like to share?
GitHub Profile
https://github.com/AddyM
GitLab Profile
No response
LinkedIn Profile
https://www.linkedin.com/in/itis-aditya-mehra/
X / Twitter Profile
No response
Personal Website / Portfolio
https://technikal.substack.com/?utm_campaign=profile_chips
If 'Other' was selected, please specify
https://medium.com/@aditya_mehra
Which PyTorch Foundation projects is the nominee familiar with?
If 'Other' was selected, please specify the project name
No response
How has the nominee contributed to the community?
If 'Other' was selected, please specify
Policy and standards work on AI safety/security
Please describe the nominee’s contributions to PyTorch Foundation projects and communities
My engagement with the PyTorch Foundation spans open-source contributions, original tooling, technical writing, community support on official platforms, conference speaking, and policy/standards advocacy. The links below back every claim.
--Open Source Contributions to PyTorch Foundation Projects
torchdiag — PyTorch model health diagnostics toolkit (creator and maintainer)
A lightweight, pip-installable diagnostic library that brings SRE-style observability to PyTorch model training: gradient health checks, dead-neuron detection, training-step verification, and memory profiling. Five commands, zero configuration, MIT-licensed.
2.GitHub: https://github.com/AddyM/torchdiag
pytorch/tutorials — Open Pull Request #3908
Authored "Visualizing the Mathematics Behind Neural Network Training," a beginner tutorial that traces how linear algebra, calculus, and optimization map directly to torch.nn.Linear, torch.autograd, and torch.optim. Every concept is verified with runnable code (torch.allclose) so learners see the math, not just read about it. CI green, awaiting review.
pytorch/pytorch — Inductor compiler diagnostics
Submitted precise root-cause analyses on two open Inductor bugs, including a verified end-to-end reproduction and a tested fix with documented tradeoffs:
PyTorch Ecosystem Contributions
3.NVIDIA NemoClaw PR #1189 — declarative host filesystem mounts for policy YAML schema
Active forks and engagement with Optuna, Opik, PyCaret (PyTorch-adjacent ML tooling)
Technical Content & Knowledge Sharing
https://technikal.substack.com/
Newsletter — Technikal (technikal.substack.com) — PyTorch series:
"Why PyTorch Isn't a Framework Anymore" — the Foundation's ecosystem consolidation story
"What Actually Happens When You Call loss.backward()" — chain rule and autograd internals
"The Compiler Wars Inside PyTorch" — torch.compile vs TorchScript vs ONNX vs TensorRT vs vLLM
"The Most Important Piece of PyTorch Engineering Nobody Talks About" — deep dive on the dispatcher
Medium technical articles:
https://medium.com/@aditya_mehra
"5 PyTorch Commands That Show You What Your Model Is Actually Doing" — practical model inspection
"Your PyTorch Model Is Using 3x More GPU Memory Than You Think" — memory profiling guide
Dev.to:
https://dev.to/aditya_mehra
"I Built a Diagnostic Toolkit for PyTorch Because I Was Tired of Guessing Why Models Fail" — torchdiag deep-dive
Grace Hopper Celebration 2025 Talk Repository:
https://github.com/AddyM/Math_behind_ML
"The Math Behind the Magic: Understanding the Role of Mathematics in Deep Learning" — slides, speaker materials, references explicitly mapped to PyTorch APIs
IEEE-USA InSight article on AI legislation in Congress (published, advocacy work bridging PyTorch ecosystem reality and federal AI policy)
https://ieeeusa.org/committees/aipc/
IEEE-USA AI Policy Committee — Co-Chair Roles & Contributions:
Co-Chair, Agentic AI Subcommittee — leading position statement
development on AI agent safety, accountability, and security
Co-Chair, Mental Health AI Working Group — drafting policy guidance
on responsible deployment of AI in mental health contexts
NIST CAISI Submission Co-Author — contributed to IEEE-USA's formal
response on agentic AI security to the NIST Center for AI
Standards and Innovation
Community Engagement on Official PyTorch Platforms
https://discuss.pytorch.org/
my profile: https://discuss.pytorch.org/u/aditya_mehra/summary
PyTorch Discuss Forum — multiple substantive technical answers, trust level promoted by the platform based on quality of contributions:
Multi-node CPU DDP with SLURM (distributed training)
MPS / torch.compile performance on Apple Silicon
libtorch 2.10 vs 2.1 CPU performance regression diagnostics
RTX 5060 Ti install troubleshooting
Conv2d batch-vs-channel shape confusion
libcudnn / libnccl link errors with prebuilt libtorch on Linux
LinkedIn Community — "PyTorch Fundamentals — Learn, Build, Contribute" Group (Founder)
https://www.linkedin.com/groups/19024016
A free learning community for PyTorch practitioners, supporting a planned weekly workshop series focused on accessibility for learners without traditional math backgrounds.
vLLM SIG CI meeting — attended; engaged on continuous integration infrastructure for the vLLM project (a PyTorch Foundation project).
Conference Speaking
Grace Hopper Celebration 2025 — "The Math Behind the Magic: Understanding the Role of Mathematics in Deep Learning" (Chicago)
PyCon US — Python community
PyCon India — Python/ML community
PyCon NZ (Kiwi PyCon) — Python community
PyCon Poland
PyCon Africa
PyCon Spain
PyOhio
ISCC (International Symposium on Cluster Computing)
Mentorship & Educational Outreach
Eagle Online Academy — Volunteer English teacher supporting Afghan women and refugees in the USA, an initiative I am exploring how to expand with accessible PyTorch fundamentals materials for first-generation ML learners.
NASA Conrad Challenge — Student mentor (high-school innovation program)
Devpost / hackathon judging — Multiple hackathons across innovation and AI tracks
ACM Peer Reviewer Certified — Reviewer for ACM Transactions on AI Security and Privacy (TAISAP)
Research, Standards & Policy Advocacy
IEEE-USA AI Policy Committee — Co-Chair of the Agentic AI subcommittee and the Mental Health AI position statement working group
IEEE P7014 Standard — Contributor
NIST CAISI Submission on agentic AI security — co-author
DOE RFI Response on AI — contributor
6 peer-reviewed publications in distributed systems and ML
2 UK Design Patents
Python Software Foundation Member
IEEE Senior Member
Primary Supporting Links
GitHub: https://github.com/addym
torchdiag (PyPI): https://pypi.org/project/torchdiag/
pytorch/tutorials PR #3908: pytorch/tutorials#3908
GHC 2025 talk repo: https://github.com/AddyM/Math_behind_ML
Newsletter: https://technikal.substack.com/
Medium: https://medium.com/@aditya_mehra
Dev.to: https://dev.to/aditya_mehra
LinkedIn: https://www.linkedin.com/in/itis-aditya-mehra/
Distinctive contribution profile: I bring an SRE/production-infrastructure perspective to the PyTorch ecosystem — building tools that automate the observability practices ML teams rarely apply, contributing technical diagnoses on the compiler stack, and creating accessible educational content that connects mathematical foundations to PyTorch's actual APIs.
Why does the nominee want to become a PyTorch Foundation Ambassador?
In 18 years of building distributed systems, I have watched technical communities succeed or fail based on a single attribute: how accessible they make their internals to people who do not arrive with the field's prerequisites. PyTorch is the rare community that built its framework around making the underlying mathematics observable — every weight printable, every gradient inspectable, every operation in the computational graph traceable. That design choice is an invitation. The community has not fully cashed it in yet, and I want to help.
My motivation rests on three threads that have converged over the past year:
Translation. When I presented "The Math Behind the Magic" at Grace Hopper Celebration 2025, the room that responded most strongly was not the experienced ML engineers — it was the career switchers, the working professionals from adjacent fields, and the students who had bounced off PyTorch tutorials that assumed three semesters of calculus. PyTorch's eager execution model makes deep learning mathematics teachable in a way no other framework does. I want to build the educational materials, workshops, and content that actually use this property — turning the framework itself into the teacher.
Production reliability. Most ML failures I have debugged in production are not novel ML problems. They are old distributed systems problems with new names: silent failures, missing observability, untested error paths. I built torchdiag because the SRE practices that took the web industry a decade to internalize have not yet reached ML teams. As an ambassador, I want to bring that perspective into the PyTorch community more systematically — through tooling, writing, and direct engagement with practitioners building real systems.
Policy bridge. As Co-Chair of the IEEE-USA AI Policy Committee subcommittees on Agentic AI and Mental Health AI, I sit at conversations where federal policy gets made about AI systems — often by people who have never written import torch. The PyTorch Foundation deserves an ambassador who can carry technical reality into those rooms and bring policy reality back into the engineering community.
An ambassador role is not a destination — it is the formal community standing that lets the writing, teaching, tooling, and advocacy work I am already doing reach further. I am committed to this work regardless of the outcome of this application. Being an ambassador would let me do it with the visibility, network, and resources to make it stick.
How would this nominee contribute as a PyTorch Foundation Ambassador?
My plan as an ambassador is to expand work I am already doing across six areas, not to start something new for the sake of the role. Each of the commitments below builds on something already in motion.
Content creation. I write a technical newsletter at technikal.substack.com with a running series on PyTorch internals — the dispatcher, autograd, the compiler stack, and the Foundation's ecosystem consolidation. I will continue this series at a steady cadence, with deeper pieces focused on production reliability, model debugging, and the parts of the framework most practitioners never look at. I will also continue publishing on Medium and Dev.to to reach developers who do not read newsletters.
Educational outreach. I founded the "PyTorch Fundamentals — Learn, Build, Contribute" LinkedIn community as a learning space for practitioners who do not come from a traditional math or research background. As an ambassador I plan to run a virtual workshop series through this group, drawing on the material from my Grace Hopper Celebration 2025 talk that mapped the mathematics of deep learning directly to PyTorch APIs. The intent is making the framework accessible to learners the typical PyTorch tutorial assumes too much from.
Organizing community events. I recently relocated to the San Francisco Bay Area, which puts me close to a concentration of PyTorch contributors, researchers, and production engineers. I would like to help organize in-person PyTorch meetups in the Bay Area — both general-purpose and focused on niches that are underrepresented at existing events, such as ML in production, model observability, and the SRE / reliability lens applied to training systems.
Mentoring contributors. I just opened my own first PR to pytorch/tutorials (#3908) and went through the workflow of forking, branching, contributing, and engaging with maintainers. I would like to help others through that same process — guiding people in my LinkedIn community and through Eagle Online Academy's network from "I have never opened a PR" to "I have an open PR on a PyTorch Foundation project." Outside the PyTorch ecosystem I have already mentored students through the NASA Conrad Challenge and judged hackathons, so this is a continuation, not a new commitment.
Technical advocacy. I built torchdiag, a small open-source diagnostic toolkit for PyTorch models (pip install torchdiag), bringing SRE-style observability practices to model training. I plan to keep developing it — adding diagnostics suggested by the community, integrating it with common training workflows, and using it as a vehicle for teaching people what to look for when their models do not learn. I will also continue submitting diagnostic comments and fixes on open PyTorch issues, particularly in the Inductor and torch.compile area where I have already started engaging.
Supporting local and global AI communities. As Co-Chair of the IEEE-USA AI Policy Committee subcommittees on Agentic AI and Mental Health AI, I am regularly in policy conversations where decisions about AI systems get made by people who have never written PyTorch code. I would like to carry the PyTorch community's technical reality into those rooms — through NIST, DOE, and IEEE submissions — and bring back what I learn into the engineering community. I have also spoken at PyCon events in multiple countries and will continue to engage with the global Python and ML community through that circuit.
The thread connecting all of this is a single commitment: making PyTorch accessible to people who would not otherwise find a way in — whether they lack the math background, the production experience, the policy context, or the geographic proximity to the existing centers of activity. An ambassador role would give that work a formal platform and a community to do it with.
Ambassador Focus Areas
If 'Other' was selected, please specify
No response
Primary Community Region
North America
If 'Other' was selected, please specify
No response
Primary Country During Ambassador Term
USA
Additional Information (Optional)
A bit of professional and supplementary context that did not fit cleanly in the contributions section but supports the case above.
Production ML background. The SRE and production-reliability lens I bring to PyTorch is grounded in actual work, not theory. I spent eleven years at VMware on ESXi / VMkernel and cloud automation, then went on to senior engineering roles at American Express, Fungible (acquired by Microsoft), Bank of America (Senior Engineering Manager, ML — leading a team of 15+ on fraud detection and retrieval-augmented systems), and Fannie Mae (Senior Software Architect, Enterprise Innovation — designing RAG and agentic AI pipelines with LangChain and LangGraph). Most of the PyTorch-in-production patterns I write about come from systems I or my teams have shipped.
Distributed systems and infrastructure foundation. Before ML became my primary focus, I worked on hypervisor and cloud infrastructure problems for over a decade. This is why the SRE lens I bring to model training feels natural — gradient health checks, dead-neuron detection, and training-step verification are the same observability practices used for monitoring service health, applied to a different layer.
Earlier open-source and community work.
Contributed to Optuna, Haystack, sklearn-pandas
Author of airegex — a small CLI tool in Go
Built BridgeAI — a WhatsApp-native AI tutor concept for refugees
Contributor to the awesome-ai-explainability curated resource list
Co-founder of Infinict (Web3/DeFi, acting CTO)
Speaking history.
PyCon US, PyCon India, PyCon NZ (Kiwi PyCon), PyCon Poland, PyCon Africa, PyCon Spain
PyOhio (regional Python conference)
Grace Hopper Celebration 2025 — The Math Behind the Magic
ISCC (International Symposium on Cluster Computing)
DC Startup & Tech Week panel on Web3 and AI
CPI Global B2B Payments Summit
Recognition and credentials not already listed above.
IEEE Senior Member (a grade requiring ten or more years of professional engineering and significant contributions to the field)
Python Software Foundation Member
ACM Peer Reviewer Certified (Transactions on AI Security and Privacy)
Two UK design patents
Approximately six peer-reviewed publications
Supporting resources.
GitHub profile: https://github.com/addym
Newsletter (Technikal): https://technikal.substack.com/
Medium: https://medium.com/@aditya_mehra (I have moved all my stories out of paywall so that anybody can read them without any payment)
Dev.to: https://dev.to/aditya_mehra
LinkedIn: https://www.linkedin.com/in/itis-aditya-mehra/
torchdiag on PyPI: https://pypi.org/project/torchdiag/
Math_behind_ML repo (GHC 2025 talk): https://github.com/AddyM/Math_behind_ML
pytorch/tutorials PR #3908: pytorch/tutorials#3908
I am happy to provide additional references, recordings, or supporting materials at the request of the review committee.