Dr. Hari Thapliyaal, PMP

Tech Stack: LLM, Prompt Engineering, Langchain, Pytorch, BERT, DistilBERT, T5, Flan-T5, BLOOM, RoBERTa, LaTeX, Mendeley, ChatGPT, ChatPDF, Python, Seaborn, Google Colab, Kaggle Kernel, Huggingface, W&B

Objective: This AI/NLP project is focused on the intricate task of generating questions, providing accurate answers, and referencing pertinent information within documents. We explored evolving technologies for generating and evaluating insightful questions, descriptive answers… Show more

Tech Stack: LLM, Prompt Engineering, Langchain, Pytorch, BERT, DistilBERT, T5, Flan-T5, BLOOM, RoBERTa, LaTeX, Mendeley, ChatGPT, ChatPDF, Python, Seaborn, Google Colab, Kaggle Kernel, Huggingface, W&B

Objective: This AI/NLP project is focused on the intricate task of generating questions, providing accurate answers, and referencing pertinent information within documents. We explored evolving technologies for generating and evaluating insightful questions, descriptive answers, determining performance metrics for question-answer pairs. We explored generating answering without giving specific context. The ultimate goal is to advance NLP capabilities in handling historical and multilingual contexts while improving the metrics of question answering systems.

Solution: In this work, we are exploring techniques for creating questions and answers for a book corpus using ChatGPT and other available techniques (QAGS). Secondly, we are finetuning the answer generation models for creating an answer (AGS). Thirdly we are retrieving a relevant document that may have answers to a given question (DRS). Finally, we are creating and evaluating a system that can answer a history question without any context (RAAGS). In this work, we used The Mahabharata book as a corpus. To evaluate the different sub-systems we have used different metrics like BLEU, ROUGE, Accuracy, Recall, R@n, P@n, F1@n, Cosine. We have used SentenceTransformer for text embedding. Our approach does not depend upon the domain, script, or language of the text document. Neither, does it depend upon the Era when the text was written. The cosine between answer & question, answer & chunk is 0.91. In DRS MRR metric is 0.55, and MAP is 0.25. In AGS cosine between the reference answer and predicted answer is 0.827. In RAAGS cosine between the reference answer and the predicted answer is 0.763


🔗 GitHub : https://github.com/dasarpai/HBQA Show less

Tech Stack: Python, PyTorch, mBERT, IndicBERT, NLP, FastText, Classification, Web Scrapping, Data Cleaning, ML, Decision Tree, XGBoost, SVC, AdaBoost, Naive Bayes, RNN, LSTM, GRU, LaTex, Mendeley, Literature Review, Research Writing

Objective : Sarcasm Detection in text written in Latin + Devanagari Script and Hindi + English + Indian
Languages (popularly called Hinglish).

Solution : Developed Hinglish Language Sarcasm Dataset. Developed custom embedding (BOW, TFIDF, Word2Vec… Show more

Tech Stack: Python, PyTorch, mBERT, IndicBERT, NLP, FastText, Classification, Web Scrapping, Data Cleaning, ML, Decision Tree, XGBoost, SVC, AdaBoost, Naive Bayes, RNN, LSTM, GRU, LaTex, Mendeley, Literature Review, Research Writing

Objective : Sarcasm Detection in text written in Latin + Devanagari Script and Hindi + English + Indian
Languages (popularly called Hinglish).

Solution : Developed Hinglish Language Sarcasm Dataset. Developed custom embedding (BOW, TFIDF, Word2Vec, fastText, Lexical), plus used embedding transfer from IndicFT, fastTextWiki, mBERT, IndicBERT. Used Task Transfer from mBERT, IndicBERT, fastTextWiki and fine-tuned the model with my training data. Developed classifier models using Logistic Regression, Support Vector Classifier, AdaBoost Classifier, XGBoost Classifier, Naive Bayes Classifier, Decision Tree Classifier, Peceptron. Developed RNN based classifier architecture. Analyzed various Embedding Models for Hinglish Text and created 108 models to identify which ML Algorithm, NLP Pretrained Model, and which embedding works best for Hinglish language and Sarcasm detection.

Key Achievement : Developed the best model with fastTextWiki embedding and Naïve Bayesian classifier gives 76% accuracy, 78% recall, 75% precision, 76% F1 score, and 80% AUC. Concluded FastTextWiki embedding Show less

- PMO Head in Dynamics NAV Implementation Project for a Major NGO. (2015)
- PMO Coach in Industrial Engineering Co. PMO Setup, defining new roles \& responsibilities, training, implementing EPM, PMO reports, and dashboards using PowerBI. (2017-2019)
- As a Head PMO defined \& implemented PMO, scorecards, KPI, and the dashboard for my Company. Implementation in Sharepoint and Microsoft Project Server. (2006-2009)
- Managed projects/programs as program/project manager, scrum master… Show more

- PMO Head in Dynamics NAV Implementation Project for a Major NGO. (2015)
- PMO Coach in Industrial Engineering Co. PMO Setup, defining new roles \& responsibilities, training, implementing EPM, PMO reports, and dashboards using PowerBI. (2017-2019)
- As a Head PMO defined \& implemented PMO, scorecards, KPI, and the dashboard for my Company. Implementation in Sharepoint and Microsoft Project Server. (2006-2009)
- Managed projects/programs as program/project manager, scrum master, and agile coach. (2005-2019)
- Consultant/MR/Lead for PMO, CMMI, ISO9001:2000 initiatives. (2007-2009)
- Delivered many training courses at international locations for the citizens of the Philippines, Singapore, Indonesia, Malaysia, Thailand, and Japan. (2013-2018)
- Classroom training experience of around 8000+ hours. (2010-Till Now)
- The Average Rating of training feedback of all training courses is 9.36 out of 10. (2010-2015)
- Managed delivery center of 100+ people.(2006-2009)
- Project Management experience of managing 30+ IT projects of 40,000 Mandays. (1996-2010)
- FPA counting experience of more than 30000 FPs of 15+ proposals/projects. (2005-2017)
- Experience at strategic \& tactical management level. (2006-Till Now)
- Experience in IT Product development lifecycle from conceptualization to selling to product improvisation. (1996-2018)
- Sound understanding of ITIL and ISO-20000, ISMS/ ISO-27001 frameworks. (2006-2009)
- A profound experience of Business and Delivery Process Engineering; defining, implementing processes, and defining metrics and quality goals for process improvement. (2005-2018)
- Solid experience in Managing OPA \& process institutionalization. Show less

- Designed PMP Course As per PMBOK Edition 4.0, 5.0, 6.0 for PMP Participants. This program is PMI, USA approved certification program. It has content of 35 hours of classroom-based training. (2010-2018)
- Designed CAPM Course As per PMBOK Edition 4.0, 5.0, 6.0 for CAPM Participants. This program is PMI, USA approved certification program. It has content of 23 hours of classroom-based training. (2010-2018)
- Designed PMI-ACP Course As per PMI-ACP course outline for PMI-ACP… Show more

- Designed PMP Course As per PMBOK Edition 4.0, 5.0, 6.0 for PMP Participants. This program is PMI, USA approved certification program. It has content of 35 hours of classroom-based training. (2010-2018)
- Designed CAPM Course As per PMBOK Edition 4.0, 5.0, 6.0 for CAPM Participants. This program is PMI, USA approved certification program. It has content of 23 hours of classroom-based training. (2010-2018)
- Designed PMI-ACP Course As per PMI-ACP course outline for PMI-ACP Participants. This program is PMI, USA approved certification program. It has content of 21 hours of classroom-based training. (2013-2018)
- Designed multiple MSP (Microsoft Project) training courses for Project Managers and other users, who want to learn this tool for project planning, governance, and reporting purposes. (2006-2018)
- Designed Dozens of other customized project management (Risk, Quality, Costing, Proposal, Vendor Management, Requirement Management, Resource Management) training courses based on the customer's need. (2010-2018)
- Designed PMP exam simulation question bank with 1000 questions. This was developed for PMBOK Edition 4.0. (2012)
- Open-house Class Room Training: Trained 4000+ leaders in 300+ corporate on PMP®, Agile Project Management, PMI-ACP®, PRINCE2, PMI-RMP®, PMI-SM®, CAPM®, Microsoft Project®, Project Online, Enterprise Microsoft Project Server®, AgileEVM, Earn Value Management, Function Point Estimation (FPA), Scrum. (2010-2018)
- Online Training: Trained 2000+ learners, 400+ sessions, 1600+ Hours. (2019-Till Now) Show less

Design and Developed dasarpAI everything from scratch for DasarpAI
Tech Stack: Jekyll, Ruby, Netlify, Github, Github Actions, Templating Engine, Markdown

website: old.dasarpai.com

Predicted driver working hours for a taxi aggregator based on seasonal, weekly, and personal behavior data.
Outcome: R² = 0.81
Tech Stack: sklearn, Python

Built a deep learning model to detect 5 hand gestures from live camera stream.
Tech Stack: CNN + RNN, Keras, Python
Outcome: Achieved 72% validation accuracy on low-resolution video feed.

Migrating dasarpAI from Jeykll to Hugo.
- Understanding strength and weakness of both technolgoies
- Understanding folder structure and difference between both
- Content migration planning
- Learning hugo's shortcode, templating engine, partial, taxonomy working
- Design entire website, menu, navigation, studied varies theme and their working
- Final migration, CI/CD, Deployment on Netlify and Github Pages

Tech Stack: Go, Hugo, NPM, Node, Netlify, Github Action… Show more

Migrating dasarpAI from Jeykll to Hugo.
- Understanding strength and weakness of both technolgoies
- Understanding folder structure and difference between both
- Content migration planning
- Learning hugo's shortcode, templating engine, partial, taxonomy working
- Design entire website, menu, navigation, studied varies theme and their working
- Final migration, CI/CD, Deployment on Netlify and Github Pages

Tech Stack: Go, Hugo, NPM, Node, Netlify, Github Action, Github, Hugo Templating Engine, Markdown

website: dasarpai.com Show less

Designed and led PMO transformations for NGOs and Industrial Firms using Microsoft Project Server, SharePoint, Power BI.

Clustered 200+ countries by health, GDP, and education metrics to recommend optimal allocation of $10M NGO fund.
Tech Stack: Python, PCA, Hierarchical Clustering, Silhouette Score