8000 joinalahmed (Joinal Ahmed) · GitHub
[go: up one dir, main page]
Skip to content
View joinalahmed's full-sized avatar
33 followers · 18 following

Achievements

Achievement: Pull Sharkx2Achievement: QuickdrawAchievement: YOLOAchievement: Arctic Code Vault Contributor

Achievements

Achievement: Pull Sharkx2Achievement: QuickdrawAchievement: YOLOAchievement: Arctic Code Vault Contributor

Highlights

Block or report joinalahmed

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
joinalahmed/README.md

Hi, I'm Joinal Ahmed 👋

Senior Product Manager - AI, Scopely · Riyadh, Saudi Arabia

🧠 Google Developer Expert in AI/ML and Google Cloud  ·  ⚡ vLLM TPU builder  ·  🎤 Speaker & Author

Previously AI Global Black Belt at Microsoft (Azure AI and Azure OpenAI) and at Google. I build and evaluate AI systems: LLM inference on TPUs, agents and their evals, explainability, document AI, and NLP for low-resource languages like Assamese. Active in the developer community in Bengaluru and beyond.

LinkedIn Google Developer Expert

What I'm working on

  • ⚡ vllm — contributing to vLLM's TPU backend for high-throughput LLM inference on Cloud TPUs.
  • 🧪 captum-on-torchtpu — Captum explainability on Cloud TPUs via TorchTPU: compatibility matrix, device-generic test suite, upstream fixes. Built for the TorchTPU Sprint.
  • 🗣️ assamese-forensic-ai — Calibrated forensic evidence for Assamese: authorship, romanisation, speaker and machine-text detection.
  • ✅ skilleval — Deterministic evaluation framework for AI agent skills. Dual-phase scoring, 82+ security patterns, no LLM judges.
  • 🧩 agentbuilder — Valogent, a visual no-code builder for multi-step AI agent workflows.

Selected projects

Area Repos
TPUs & inference vllm · torch_tpu · captum-on-torchtpu
Agents & evals skilleval · agent-eval · agentbuilder · mcp-client
Document AI autoprocure · invoiceparsingwithAOAI · reimbursement-app-AOAI
RAG & enterprise AI RAGwithRBAC · PrivateGPT · AgenticHRSystem · chat-ui
On-device & speech ondevice-slm · azurespeech · spoken-control-panel
MLOps abalone-model-build · abalone-model-deploy

Community

  • Google Developer Expert, AI/ML and Google Cloud
  • AI Converge
  • TensorFlow User Group Bengaluru
  • Google Developer Group Bengaluru

Stack

Python PyTorch JAX TPU vLLM TypeScript Google Cloud Azure AWS

GitHub stats

Pinned Loading

  1. assamese-forensic-ai assamese-forensic-ai Public

    Forensic AI for Assamese: calibrated evidence for authorship, romanisation, speaker and machine-text detection. Reproduces four CPU-only experiments with EER, Cllr and Tippett plots.

    Python

  2. captum-on-torchtpu captum-on-torchtpu Public

    Captum explainability on Cloud TPUs via TorchTPU: compatibility matrix, device-generic test suite, upstream fixes and a tutorial. Built for the TorchTPU Sprint.

    Python

  3. skilleval skilleval Public

    100% deterministic evaluation framework for AI agent skills with dual-phase scoring, 82+ security patterns and transparent grading. No LLM judges.

    Python

  4. AgenticHRSystem AgenticHRSystem Public

    StaffAssist: AI-powered resume analysis for HR. Upload resumes and ask natural-language questions, with evidence-backed answers from Azure OpenAI Assistants and FastAPI.

    HTML 2 1

  5. agent-eval agent-eval Public

    Evaluation harness for a customer service AI agent on Google Cloud: test cases, scoring and reporting.

    Python

  6. agentbuilder agentbuilder Public

    Valogent: a visual, no-code AI agent workflow builder. Draw multi-step agent flows on a canvas, configure nodes, and run them live.

    JavaScript

0