← Back to cohort

Khizar Shabir

FAST · 2020 · i16 - 0294

—

Phone

—

GitHub

—

Academic

Program

BSCS

CGPA

—

Year

2020

Education

SEECS

Address

—

DOB

—

Career

Current role

—

Target role

—

Skills

PyTorch, Python, TensorFlow, React

Verbatim text

The exact text the LLM saw on the page (or the booklet text from the old import). This is what powers semantic search.

NEO – AI Conversational Agent 
 
A conversational expert for chit-chat, having its conversation consistent with its persona. Absence 
of consistency in a conversational agent has been a long-standing problem. Conversation 
consistency is very important if a conversational agent wants to build trust and long-term 
confidence. Personality in a key differentiator in a conversational agent. We have worked on 
improving the conversation consistency of the agent according to its persona. 
 We trained the language model on movie subtitles data set with the use of word embedding and 
transfer learning techniques in Python. We used personality sentences, history of conversation as the input context to the language model. We have used a next step prediction loss and language model loss and added them together. After fine-tuning the model on our dataset. The model was 
able to predict the next sentence tokens. 
We have used the top-k and top-p sampling as a decoder mechanism. Top k sampling keeps the token that fall in certain range of probability and discard other tokens. It helps the model from going 
of topic. Trained dialogue model is able to learn its personality and capture the context of the dialogue.

AI enrichment

Khizar Shabir holds a BSCS degree and has experience developing AI conversational agents using Python, word embeddings, and transfer learning. The candidate has worked on improving conversation consistency and personality retention by fine-tuning language models on movie subtitle datasets.

Skills (AI)

["Python", "Natural Language Processing", "Transfer Learning", "Word Embeddings", "Language Model Fine-tuning", "Top-k Sampling", "Top-p Sampling", "Dialogue Systems"]

Status: ai_done

Provenance

Source file: Graduate Directory FAST School of Computing 2020 (Final Complete) (1).pdf
From job #23 page 209
Created: 1778170703