Deepak Victor

Machine learning engineer

About Me

Passionate about developing deep neural networks for computer vision and natural language processing with a focus on researching and fine-tuning state-of-the-art models for unstructured data.

Experience

Rapid acceleration partners

Machine learning intern

July 2023 - Jan2024

https://www.linkedin.com/company/rapid-acceleration-partners/

Practical AI Solutions for Digital Business Transformation.

worked as a Machine learning engineer intern on hyper-automation by processing unstructured data with artificial intelligence. Fine-tuned real-time document data using large language models, such as Phi-2 and Mistral.Worked on visual language models, document extraction and object detection.

Trycom AI

Machine learning Engineer

Jan 2024 - present

https://trycom.ai/

TRYCOM AI : Simplifying content marketing with AI precision.

As a Machine Learning Engineer , I focus on automating branded and personalized content creation by fine-tuning LLMs for specific tasks, ensuring content is human-like and SEO optimized.Also i work on the process of Converting these deep learning models to onnxruntime and implementing,integrating,dockerizing and deploying these codes.

Projects

Techbytes AI

Instagram Page

An Ai Bot that Tracks the Recent information (For every six hours) about a given Topic and Creates a Insta or Fb or Twitter Post with Image and uploads it using the Platforms API.

This project is right now hosted on ec2 and implemented with the help of Groq, Perpleixity discover (right now), Scrapeops API’s. The post are created using a smart Image-Text embedding program.The live demo for this project is running successfully on techbytes insta page.These posts are uploaded for every six hours using instagrams content publishing API with AI summarized content and hashtags.

Perplexity Clone

Github page

Perplexity Clone that could use Groq's LPU power and could work as your Upto date knowledged AI assistant

This project Focuses on scraping the top google results or you Brave’s search api to get top results and by Scraping the Articles we could provide the LLM recent knowledge about things happening and can be questioned about recent incidents.Groq API is being used for LLM inference and Nomic embedding is used for RAG implementations. Qdrant vector database is used to store the vector embeddings and retrieve it faster.

Moviesense

Website Link

An Ai system that suggests movies related to the given prompt. this system returns results from 4 lakh movies in milliseconds in a 1GB RAM system.

This project is built using customized sentence transformers. nearly 4 lakh movies are optimized embedded and stored in a vector database. this system is hosted in a docker service in AWS free ec2 instance of 1GB ram. systems more than 4GB ram can run a higher accuracy ouput with rag implementation.

Tamil Bert

Model link

this model is a bert model trained as Masked language modelling task which can be used for many tasks like clasification, named entity recognition etc.

The model and the tokenizer is trained from scratch for the language tamil. this model is trained with 10.6 Million tamil sentences on a p100 for 18 hrs resulting in a better evaluation score for other tasks.

nerf Neural radiance field is a technology that converts a bunch of 2D images to 3D model.

using the original implementation in tensorflow i reassembled the code for pytorch. this allows to use customized features that pytorch offers.

SVCE-BUS

a progressive web app which tracks the location of the college bus without the help of gps

this webapp is developed on react js and django uses the drivers mobile to track the drivers location and sents the coordinates to the students. this uses leaflet open source map framework to display the coordinates. this system was professional with security features and login systems.

OCR_From_Scratch

Github Link

OCR for language tamil

Optical Character Recognition (OCR) is the technology that allows the conversion of printed or handwritten text into digital text. It has been widely used for various applications, including language translation, text mining, and document digitization. Tamil OCR specifically focuses on recognizing and extracting text from documents written in the Tamil language.

FACIT

webapp

A simple classification system that classifies gender based on the image

Developed on convolutional layers with the data scrapped from google images

Obscene Detector

Obscene detector that detects NSFW image

this system uses masktransformer to detect the human outline of the image and then uses a classification model to classify.

wikipedia datascrapping

Scrapped 36000 movies plot and 2000 songs from various websites including wikipedia

gained experience by datascrapping many websites for self supervised training

Education

Sri venkateswara college of engineering

B.tech Artificial intelligence and Data science

2020 - 2024

Pursued my degree on artificial intelligence and data science with a CGPA of 8.59. Devloped interest on deep neural networks and data processing here.

A Little More About Me

i like to watch movies a lot and play lots of games. this stage developed my interest towards computers. within a short period of time i tried all the skills that are related to computers.developing games, cybersecurity, web development, blockchain, automation which ended up on developing neural networks to solve real word problems.