Skip to content
View Harras3's full-sized avatar

Block or report Harras3

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.

Maximum 250 characters. Please don't include any personal information such as legal names or email addresses. Markdown supported. This note will be visible to only you.
Report abuse

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

Report abuse
Harras3/README.md

Hi there, I'm Harras Mansoor πŸ‘‹

AI Engineer | Building Production-Scale AI Systems | Open Source Contributor

I specialize in developing and deploying advanced AI systems with a focus on RAG, LLMs, Voice Agents, Speech-to-Text, and Text-to-Speech. Currently engineering AI solutions for Fortune 100 companies at Avirso.

πŸ”¬ What I'm Working On

  • Contributing to Hugging Face TRL (Transformer Reinforcement Learning) - 16.3k+ ⭐
  • Contributing to OpenVoiceChat - A library for creating voice agents (252 ⭐)
  • Researching hallucination-free speech-to-text systems
  • Developing end-to-end audio models for voice interactions

🎯 Notable Projects

πŸ—£οΈ Unhallucinated Faster Whisper - Reducing hallucinations in OpenAI Whisper models when processing audio with human noise. Available on PyPI and actively used in production environments.

πŸ€– OpenVoiceChat - Open-source library enabling developers to build sophisticated voice agents with integrated RAG capabilities and end-to-end audio processing.

βš–οΈ CaseLink - AI-powered all-in-one solution for law firms featuring legal research assistant with RAG, document generation, and Redis caching for optimized performance.

πŸ”Š First Arabic EOU Model - Created the first open-source end-of-utterance detection model for Arabic language, enabling natural conversation flow in voice agents.

⚑ Enterprise-Grade RAG - Modular RAG implementation with NVIDIA NeMo guardrails, semantic caching, and Redis vector database.

πŸ“š Published Research

Mitigating Hallucinations in Speech-to-Text Systems - IEEE 4th International Conference on Computing and Machine Intelligence (ICMI)

πŸ’» Tech Stack

AI/ML: PyTorch β€’ TensorFlow β€’ LangChain β€’ LangGraph β€’ CrewAI β€’ Hugging Face

Infrastructure: CUDA β€’ TensorRT β€’ Triton Inference Server β€’ Microsoft Foundry β€’ Google Vertex AI

Databases: Redis β€’ Weaviate β€’ MongoDB β€’ DynamoDB β€’ Google Bigtable

Languages: Python β€’ C++ β€’ C β€’ C# β€’ CUDA β€’ Go

πŸ† Achievements

  • 3 merged PRs to Hugging Face's TRL repository
  • Published research paper at IEEE ICMI conference
  • 2x AWS Scholarship recipient for Nanodegrees
  • President of Microsoft Learn Student Ambassadors at FAST NUCES Islamabad

πŸŽ“ Certifications

  • Model Parallelism: Building and Deploying Large Neural Networks - NVIDIA
  • Custom ASR for Speech AI - NVIDIA
  • Computer Vision for Industrial Inspection - NVIDIA

πŸ“« Let's Connect

LinkedIn Email


πŸ’‘ Passionate about pushing the boundaries of AI, particularly in voice and speech technologies. Open to collaborations on cutting-edge AI projects.

Pinned Loading

  1. unhallucinated-faster-whisper unhallucinated-faster-whisper Public

    'unhallucinated-faster-whisper,' a powerful enhancement built on the faster-whisper framework for audio transcription.

    Python 10 1

  2. Enterprise-Grade-RAG Enterprise-Grade-RAG Public

    This is a RAG based chatbot in which semantic cache and guardrails have been incorporated.

    HTML 13 3

  3. trl trl Public

    Forked from huggingface/trl

    Train transformer language models with reinforcement learning.

    Python

  4. CUDA-Project CUDA-Project Public

    Developed CUDA-optimized code for efficient training of deep learning models.

    Cuda

  5. Arabic_EOU Arabic_EOU Public

    End of utterance model for Arabic

    Jupyter Notebook

  6. Multithreaded-dummy-Neural-Network Multithreaded-dummy-Neural-Network Public

    This is code for a multithreaded dummy neural network in C++

    C++