Hassan Javed

AI Researcher

InkAI • Período integral

Oct 2024 - Present • 1 yr 7 mos

A startup founded by Rich Miner, Co-founder of Android. Working at the cutting edge of Artificial Intelligence for digital handwriting synthesis and intelligent search. As a Deep Learning & NLP Engineer, I built production-grade AI systems across the full Machine Learning lifecycle: ▸ Transformer & Deep Learning: Implemented TrInk (EMNLP 2025) a state-of-the-art handwriting generation model featuring MDN output layers, cross-attention multi-writer style conditioning, Gaussian memory masks, and polar coordinate tokenization. Researched adaptive Bézier curve segmentation as an alternative tokenization strategy with continuous 10D feature vectors for transformer inputs. ▸ TensorRT Optimization: Converted the PyTorch model to ONNX + TensorRT FP16, achieving ~2× GPU inference speedup on NVIDIA RTX 4500 Ada. Diagnosed and resolved FP16 numerical precision instabilities in autoregressive generation pipelines — real-world Deep Learning deployment engineering. ▸ ML Model Serving: Built a production FastAPI server on GCP with PostgreSQL LRU caching, adaptive batch GPU processing, and word-length-based step scheduling. Fully integrated with Android and web clients via RESTful APIs — complete Machine Learning DevOps lifecycle. ▸ NLP & LLM Engineering: Developed an AI-powered semantic search system using ChromaDB vector database and OpenAI LLMs for intelligent, context-aware notebook content retrieval — applied NLP and LLM engineering at the product level. ▸ Data Science Pipeline: Engineered a large-scale Amazon Mechanical Turk (AMT) data collection system with boto3 automation, Google Drive/Sheets tracking, and multi-phase worker recruitment for curating a handwriting dataset at scale. Artificial Intelligence | AI Developer | Deep Learning | Machine Learning | LLM Engineer | NLP | Data Scientist | AI Chatbot | TensorRT | FastAPI | GCP | Transformer | Vector Database | MLOps

AI Developer

NASTP • Período integral

Mar 2024 - Present • 2 yrs 2 mos

Deployed and engineered large-scale, production-grade Artificial Intelligence and Machine Learning systems in a high-security, real-world environment: ▸ LLM Infrastructure & Scalable Serving: Deployed and managed a 5-server vLLM cluster (NVIDIA RTX 4500 Ada) serving Llama 3.1 8B with PagedAttention KV cache optimization, FP8 quantization, prefix caching, and Prometheus/Grafana monitoring — enterprise-scale LLM Engineer work in production. ▸ Distributed Multi-GPU AI: Implemented distributed inference for LLaMA 3.2 70B across 4 NVIDIA GPUs using distributed-llama.cpp and vLLM with Layer 3 load balancing for real-time streaming token generation. ▸ AI Chatbot & RAG System: Architected a full-stack FastAPI RAG AI Chatbot with Qdrant vector database, vLLM backend, cross-encoder reranking, RAGAS + F1/EM evaluation, JWT auth, and a React frontend with conversation management. Improved answer accuracy from 70% to 80%+ in an air-gapped, offline environment. ▸ LLM Fine-Tuning & NLP: Fine-tuned LLaMA 3.2, DeepSeek, Qwen2.5, and GPT using QLoRA on custom PDF datasets for domain-specific NLP question-answering. Deployed 70B GGUF models offline via llama.cpp on A100 GPUs with Open WebUI. ▸ Deep Learning & Computer Vision: Trained YOLOv8/v9 for real-time object detection; built LSTM models on 1M-record synthetic datasets for time series forecasting (75%→89% accuracy); applied SHAP explainability to anomaly detection for defense systems. ▸ Data Science & Geospatial AI: Engineered hybrid LangChain + LangGraph pipelines combining vector and graph databases; built geospatial ML pipelines using QGIS for flight path simulation. Artificial Intelligence | LLM Engineer | AI Chatbot | NLP | Machine Learning | Deep Learning | Data Scientist | AI Developer | RAG | Fine-Tuning | Computer Vision | vLLM | Vector Database | MLOps

AI Research Executive & NLP Engineer

PanaceaLogics • Período integral

Jul 2023 - Mar 2024 • 8 mos

Leading end-to-end Artificial Intelligence and Machine Learning projects across NLP, LLM engineering, and Computer Vision from proof-of-concept to production. ▸ NLP & Document Intelligence: Implemented advanced NLP solutions using LLMs, LayoutLMv2/v3, LiLTv2, OCR pipelines, and GPT-2; built AI chatbots and intelligent document extraction systems powered by BERT-based models and LangChain production-ready NLP engineering. ▸ LLM Fine-Tuning & Optimization: Applied RAG techniques to LLaMA 2/3 models; optimized performance using LoRA and QLoRA fine-tuning methods; built OpenAI API-powered proof-of-concept NLP applications with Flask for domain-specific question-answering. ▸ PDF-Based AI Chatbot & Semantic Search: Developed a query-based semantic retrieval system across uploaded PDFs an early-stage AI chatbot for intelligent document Q&A, combining NLP and vector-based search. ▸ Computer Vision: Collected, labeled, and preprocessed fish datasets for Deep Learning classification, segmentation, and object detection pipelines; implemented fish BMI estimation via image analysis and stereo-vision-based monitoring systems. ▸ Data Science & Machine Learning: Built and validated end-to-end ML pipelines covering data collection, preprocessing, augmentation, model training, and evaluation for domain-specific AI applications. ▸ Team Leadership: Mentored junior AI Developers and interns in data annotation, labeling, and Deep Learning workflows driving knowledge transfer across multidisciplinary teams. Artificial Intelligence | NLP | LLM Engineer | AI Chatbot | Machine Learning | Deep Learning | AI Developer | Data Scientist | Computer Vision | RAG | Fine-Tuning | LangChain | Document AI | Semantic Search