Meghana Bhange
Verified Expert in Engineering
Software Developer
Meghana是一名机器学习工程师,热衷于以数据驱动的方式解决问题. 她目前正在可信信息系统实验室攻读硕士学位,研究重点是隐私保护技术, ÉTS Montréal. 她在自然语言处理方面有经验,之前曾在SemEval-2020上发表过作品. Meghana热衷于从事创造性项目,并总是寻找新的方法来应用她的技能.
Portfolio
Experience
Availability
Preferred Environment
生成预训练变形器(GPT),自然语言处理(NLP),人工智能设计
The most amazing...
...我参与的项目是开发一个端到端的自定义识别服务,该服务基于资源受限的代码混合设置,具有低延迟需求.
Work Experience
AI Engineer
UInclude, Inc
- 利用SpaCy和基于规则的引擎开发了一个特定于上下文的偏倚词匹配模型,以识别工作列表中的偏倚词.
- 使用句子转换器和GPT-3创建了同义词扩充器,以发现上下文特定的同义词, 用无偏见的替代词取代工作列表中有偏见的词.
- 使用AWS上的FastAPI端点部署模型,同时通过DynamoDB存储和查询数据.
Researcher
TISL Lab at ETS Montreal
- 研究了一个投诉申诉系统的隐私保护ML和数据发布.
- 研究了基于反事实解释api的机器学习系统的模型提取攻击.
- 建模一个可以利用反事实解释提供的信息来构建高保真度和高精度模型提取攻击的对手.
- Benchmarked the model performance on the Folktables dataset, with the extracted model gaining fidelity of around 97.6%.
OpenAI Developer
Zurney.app
- 构建一个集成了GPT-3 API的FastAPI后端,生成旅行行程并提取地点. These locations were then geo-encoded with co-ordinates.
- Built a Next.js应用程序显示旅行行程,并显示谷歌地图上的地理位置颜色代码对应的日子在旅行和有关每个位置的信息.
- Dockerized and deployed both the FastAPI back end and Next.js front end to DigitalOcean.
Machine Learning Engineer
Hunters.ai
- 研究并构建用于评估威胁搜索检测器和理解检测输出中的异常模式的分析工具.
- 组织机器学习检测器的监控和质量检查基础设施.
- Created a framework for deep investigation of threats.
Machine Learning Engineer
The Verloop.io
- 对意图识别服务做出贡献,使用句子转换器提高top-K的查全率和准确率, which improved F1 by 40% absolute.
- 设计、构建并部署跨所有客户端的多语言名称识别服务.
- 评估各种语言模型(如ULMFiT和VAMPIRE)在低资源语言上下文中的性能.
- 使用生成式预训练变压器3 (GPT3)人工智能在聊天机器人中为FAQ系统创建合成训练数据.
Machine Learning Intern
The Verloop.io
- 创建了为多语言对话定制的人名提取器. Tweaked Flair, Facebook's natural language processing library, to work on low-latency use cases in English, Spanish, and French.
- 与之前部署的FastText模式相比,最终模型在F1中的效率提高了47%.
- 将开发的多语言名称提取器部署到生产环境中,总延迟低于500毫秒.
Experience
Model Extraction Attack Using Counterfactual Explanation
LitNER | Literature Named Entity Recognition
http://github.com/meghanabhange/litNERHinglish Twitter Sentiment Detection | SemEval2020
http://arxiv.org/abs/2008.09820Wikipedia Textbook Assistant
http://github.com/meghanabhange/Wikipedia-Textbook-Assistant人工精神错乱(稳定扩散反人类卡牌)| total Hackathon
我根据DALLE和Stable Diffusion的质量和延迟对性能进行了基准测试. Also, 我在FastAPI上部署了最终的模型,使其更容易与后端的其余部分集成. The solution won the second prize in the Hackathon.
Skills
Languages
Python, SQL, Python 3
Other
Machine Learning, Natural Language Processing (NLP), Artificial Intelligence (AI), Deep Learning, APIs, Text Generation, Language Models, GPT, Data Engineering, Chatbots, OpenAI, AI Design, Machine Learning Operations (MLOps), Large Language Models (LLMs), Computational Linguistics, Generative Pre-trained Transformers (GPT), OpenAI GPT-3 API, Research, Streamlit, Transfer Learning, BERT, Signals, Information Theory, Custom BERT, Stable Diffusion, DALL-E, FastAPI, Inference API, Speech Recognition, Web Development, DaVinci, Systems, ChatGPT, Cryptography, Information Technology, Prompt Engineering
Frameworks
Django, Flask, Next.js
Libraries/APIs
Pandas, Scikit-learn, SpaCy, TensorFlow
Storage
数据管道,PostgreSQL, Amazon S3 (AWS S3), Amazon DynamoDB, Google Cloud
Tools
Slack, Named-entity Recognition (NER), Amazon SageMaker
Platforms
Kubernetes, Google Cloud Platform (GCP), Amazon Web Services (AWS), Visual Studio Code (VS Code), DigitalOcean, AWS Lambda
Industry Expertise
Cybersecurity
Education
Master's Degree (Ongoing) in Information Technology Engineering
École de Technologie Supérieure - Montreal, Canada
电子与通信工程专业本科以上学历
Savitribai Phule Pune University - Pune, India