Hello! I am Apurva Thakur
Apurva Thakur
Techstack
Docker
MongoDB
Azure Devops
Azure
Redis
Javascript
Python
AWS
Git
Experience
AI/ML Engineer
AI/ML Engineer
- Developed and deployed ML pipelines on Azure with CI/CD integration using Azure DevOps and GitHub Actions for streamlined model lifecycle management.
- Built multimodal GenAI workflows using GPT models, LangChain, FastAPI, and Deepgram to automate tasks involving both text and audio inputs.
Data Engineering Apprenticeship
Data Engineering Apprenticeship
- Led a 20+ member team in prompt engineering and annotation tasks for over 3 LLM-based projects, ensuring high-quality data pipelines for GenAI systems.
- Designed and prototyped frontend UI using React.js and Figma for the company’s in-house VZsmart GenAI platform.
Recent work
1. Vizismart – Multimodal GenAI Assistant
Key techstack
- YOLOv8
- GPT-4o
- Deepgram
- LangChain
- FastAPI
- Docker
- AWS S3
Overview
Vizismart is a real-time, multimodal GenAI assistant that combines computer vision and natural language understanding. It leverages YOLOv8 and GPT-4o to interpret both visual and voice inputs, delivering intelligent, contextual responses for dynamic interactions.
Key Features
- YOLOv8-Based Vision: Trained on 1000+ annotated images, achieving 92% accuracy in real-time object detection tasks.
- Multimodal Input Handling: Processes voice and visual inputs using Deepgram for speech-to-text and LangChain for GPT-4o routing.
- Scalable Deployment: Dockerized with persistent logging and monitoring via AWS S3 for reliable, production-grade performance.
- Fast & Context-Aware Responses: Integrated Groq for ultra-low-latency inference, enhancing user experience.
2. Azure MLOps Pipeline
Key techstack
- Azure ML
- Azure DevOps
- Python
- scikit-learn
- Random Forest
- CI/CD
Overview
A robust, production-grade MLOps pipeline deployed on Azure, featuring automated training, CI/CD, and advanced model governance. Designed for continuous improvement and seamless integration into enterprise workflows.
Key Features
- Automated Model Retraining: Integrated Random Forest model (87% F1-score) with CI/CD pipelines using Azure DevOps.
- Dataset Versioning & Reproducibility: Ensured traceable experiments and consistent model outputs with Azure ML tracking.
- Drift Detection & Rollbacks: Implemented automatic data drift monitoring and rollback mechanisms for reliability.
- Enterprise-Ready CI/CD: Enabled continuous integration and delivery with modular pipeline components and version control.
3. Job-Hunting Automation System
Key techstack
- Apify
- OpenAI API
- Telegram Bot
- Google Sheets API
- Node.js
- Prompt Engineering
Overview
An end-to-end job search automation tool that scrapes listings, matches jobs to resumes using AI, and auto-applies with personalized cover letters—all while keeping logs and alerts updated in real-time.
Key Features
- Automated Job Scraping: Extracted 500+ job listings per week using Apify for continuous opportunity discovery.
- AI-Powered Job Matching: Used OpenAI’s resume-job match API to score job relevance and prioritize top listings.
- Auto Application via Telegram Bot: Seamless application submission with interactive Telegram-based UI.
- Personalized Cover Letters: Dynamically generated emails and cover letters using prompt templates for 100% tailored outreach.
4. Yumbot – Food Delivery Chatbot
Key techstack
- Dialogflow
- FastAPI
- MySQL
- Python
- Natural Language Processing
Overview
Yumbot is an AI-powered chatbot for food delivery services, designed to streamline order placement and reduce customer support workload. It uses natural language understanding to handle diverse user intents with high accuracy.
Key Features
- Conversational AI with Dialogflow: Handled 30+ food ordering intents with 95% accuracy using advanced NLU and NER.
- Backend Integration: FastAPI server connected with MySQL database for real-time menu, order, and user data handling.
- Customer Support Automation: Reduced manual service efforts by 60% through smart query handling and response generation.
- Scalable Architecture: Modular backend ready for integration with POS and delivery APIs.