Amgad Hasan

Amgad Hasan

Machine Learning Engineer

AI-powered products and services, specializing in NLP and STT applications.

Services

Strategic Consultation

Helping teams define problems, validate ideas, plan actions, avoid pitfalls, and set up ML best practices.

  • Defining the problem properly
  • Validating your idea and the potential product market fit (PMF)
  • Planning the best course of action to achieve your goals
  • Avoid common pitfalls that cost your small, expensive team weeks of extra work
  • Set up machine learning & MLOps best practices for your team to reduce technical debt over the long term
Problem Definition Idea Validation Product Market Fit Action Planning Pitfall Avoidance MLOps

Tactical Implementation

Selecting models, deploying LLMs and STT models, fine-tuning, custom evaluation, and high-quality documentation.

  • Selecting the right model and pipeline for your use case
  • Deploying LLMs and STT models at scale to serve thousands of users efficiently
  • Implement retrieval augmented generation (RAG) pipelines to add knowledge to your chatbots
  • Fine-tuning on your own data to improve the accuracy and reduce hallucination
  • Develop custom evaluation pipelines to assess and monitor models accuracy
  • Write high-quality documentation, blog posts and architecture diagrams
Model Selection Deployment Fine-tuning RAG Evals Documentation

Technical Writing and Content Creation

Technical storyteller and perpetual learner. I transform complex concepts into compelling content that developers love. Whether you need crystal-clear documentation, engaging blog posts, or user-friendly guides, I help tech companies turn technical writing into a powerful marketing tool. Let's elevate your product's narrative and connect with your audience.
Check some of my work below:

Model Selection Deployment Fine-tuning Custom Evaluation Documentation

Projects

Sera Agent 0.1

An open source cybersecurity AI assistant

  • Integrated with Sera
  • Curated trainig data set for supervised finetuning (SFT)
  • Optimized fine-tuning code to utilize hardware efficiently, costing <500$
  • Open source checkpoint
  • Built on top of LLama 3
NLP Agents LLMs Finetuning SFT Function Calling Tool use

Improving transcription accuracy

Built a custom speech-to-text model for a clinical documentation service that offer ambient note-taking.

  • Curated and prepared +700 hours of audio
  • Efficiently trained model on budget hardware utilizing distributed training
  • Increased trascription accuracy by 30%
  • Helped the company raise $425,000 in a pre-seed round
LLMs Chatbots Quantization Deployment Speculative Decoding Latency

Optimizing LLM Latency

The goal was to significantly icrease the output speed of NeuralDaredevil-7B for copywriting uses.

  • Generated multiple quantized versions
  • Setup an eval pipeline for quantized versions to maintain model output quality
  • Used speculative decoding to fully utilize the compute capability of high-end gpus
  • Achieved inter-token latency of 7 milliseconds per token for a 7B model on 1xA100
LLMs Chatbots Quantization Deployment Speculative Decoding Latency

Recorded Talks

Whisper v3 Turbo

A talk I gave in the Latent Space LLM Paper Club about the newly released Whisper Large v3 Turbo

GPT-1

A talk I gave in the Latent Space LLM Paper Club Asia Edition about the GPT-1 model from OpenAI

Contact Me

Get In Touch

Ready to collaborate? Book a free 30-minute consultation to explore challenges and develop solutions that drive your project’s success.

Email Me