人工智能求职

Designed and developed a chatbot system for customer service using C++ based on TensorFlow framework

Collected 120GB open domain question answering texts and performed ETL with building end-to-end pipeline

Implemented word embeddings and applied NLTK, spacy to for text tokenization, parsing and classification

Reproduced and trained the modern BERT and Transformer based models, benchmarked with OpenAI GPT-3 for performance evaluation

Designed distributed system to architect the pipeline infrastructure to expectedly deploy the models online

Automated the compiling, building and deployment process with CI/CD

Designed and built a transformer-based neural network for video retrieval, leveraging CLIP and TimeSformer to enable semantic search over 100M+ video clips.

Constructed a distributed data processing pipeline using Apache Beam and TensorFlow DataService, reducing preprocessing time by 80%.

Applied low-bit quantization and knowledge distillation to deploy efficient models on edge devices, improving inference speed by 5x.

Implemented video embedding caching using Faiss and ANN search, reducing query latency from 2.3s to 200ms.

Integrated a multi-modal search system combining text, audio, and video embeddings to improve retrieval relevance by 32%.

Deployed a cloud-native pipeline with Kubernetes, Terraform, and ArgoCD, enabling CI/CD for frequent model updates with zero downtime.

Developed a system for verifying adversarial robustness of neural networks on GPUs, scalable to millions of parameters.

Built a graph parser for neural networks and optimized the computation kernels using JIT compiler and TorchScript.

Devised a super gradient descent method in PyTorch enabling recursively optimizing layer-wise per-neuron constraints.

Proposed a novel cost model in Python for the branch-and-bound optimization algorithm, resulting in 1.5x speedup.

Benchmarked on 12 diverse verification tasks and achieved 98.1% accuracy, ranking it as the state-of-the-art algorithm.