Akhilesh Vijayan

Resume parser or AI Assistant

• Tech stack used - Fast API, finetuning NER BERT models, finetuning detectron2 (Faster CNN), AWS S3 bucket, PostgreSQL database, OpenAI (finetuning and prompt engineering), FastText model training. ReactJS, Azure deployment.

• As an AI developer, I single-handedly developed the resume parser project from scratch and successfully released it to production within a year.

• Led a team of 10 data preprocessing analysts to support NER and section layout model training based on data annotated using Label Studio.

• Completed the first phase using Python Flask framework and converted it to FastAPI to enable asynchronous processing and to improve the performance.

• Completed numerous POCs as a part of this research project such as Detectron2 layout detection with 1000 training data, NER model training, and finetuning using different BERT NER models, finetuning models using Optuna to optimize the model to avoid overfitting.

• Created a resume viewer using ReactJS as frontend and Python FastAPI framework as backend enabling data access to the interns without providing them the physical data to ensure data privacy.

Smart candidate search

• Tech stack used: Elastic search, BERT model (paraphrase-MiniLM-L6-v2) for vector embedding, Django.

• Developed a smart candidate search to search the candidates with their skills, job titles, years of experience, and location using vector similarity search between the query and the sentence created using the candidate’s data by prioritizing the abovementioned four parameters.

• Candidate’s index was created in Elasticsearch and a new key called “vector” was added by creating vector embedding using the BERT model. Vector has been stored and made use of vector similarity search in Elasticsearch. Checking the similarity of the query vector and sentence vector helps us provide the most similar data on the top.

• As phase 2 of the project, used OpenAI’s speech recognition model (Whisper) to convert voice to speech to integrate it into the smart search so that recruiters can either provide text or voice as input to the search engine.

• MYSQL query development for data extraction
• Worked on Drain3 to extract log templates from system log data and FastText, Spacy, and Genism to generate word embeddings using Word2vec
• Worked on different ML models to predict anomalies from system logs
• Exploratory Data Analysis of the system log data to get predictive insights and take future mitigation methods
• I used the Boto3 module in Python to extract the logs from Amazon S3 Bucket.

• Handled 40+ calls and 150+ emails daily from UK-based clients reporting their application-related issues
• Used Bomgar as remote access to fix application-related issues within the call itself
• Incident management and user management
• Handled incident calls affecting 1000+ customers within the call itself by dealing with it efficiently
• MySQL query development and execution.
• Managed Virtualization Environment using VM Ware ESXI 6.7 & Virtual Center for Development
• Azure & VMware VDI and Server Creation for Clients
• Creation of Amazon workspace using Jenkins and Azure VM to provision workspace access for Customers
• Maintaining user accounts and permissions using IAM
• Experience in ticketing tools such as Service Now, Remedy on Demand, and JIRA.

• Monitoring, implementing, and configuring network devices and troubleshooting network-related problems
• Handled network-related calls and efficiently troubleshoot issues to solve network issues on client sites with the help of remote desktop and putty.