Blogs

Rakesh Bhugra

I am a Full Stack
AI Engineer

GitHub Contributions

Activity on @rakeshbhugra

 

Experience

Vurbalize

AI Engineer (2021 – Present)

Post-GenAI Pivot Phase

Vurbalize Conversation Engine (VCE) & AI Systems
  • • Architected and deployed a multi-tenant, enterprise-scale AI conversation platform (VCE) on Microsoft Azure, serving over 4 major B2B clients and implementing role-based access within a microservices architecture.
  • • Developed a multi-LLM orchestration system with fallbacks and safety guardrails (toxicity & manipulation checkers) that handled PII (Personal Identifiable Information), and architected a Hybrid RAG pipeline using a custom reranker over knowledge graph, full-text, and semantic search results.
  • • Implemented a comprehensive evaluation framework, including an LLM-based automated scoring system, a human-in-the-loop testing pipeline, and support for structured red-teaming exercises to ensure system robustness.
  • • Built automated website crawling modules to ingest and process client data, continuously populating the knowledge base for the RAG system.
Agentic Research & Automation
  • • Engineered a multi-agent AI system using LangGraph to automate and significantly accelerate autonomous prospect research.
  • • Architected deep research agents that leverage multi-step reasoning to perform targeted web scraping, LinkedIn data extraction, and execute queries via a programmable Google search engine, while also communicating with internal AI systems for analysis.
  • • Implemented stateful, long-running research workflows using LangGraph for persistent state management and high reliability.
Live Agent Portal
  • • Built an enterprise-grade, real-time customer support platform from the ground up using Next.js/React, FastAPI, and Socket.IO.
  • • Engineered a live browsing portal allowing agents to monitor user website activity in real-time, featuring a proprietary algorithm to rank and surface high-intent users for proactive engagement.
  • • Integrated a WebRTC-based video calling system using the Twilio Video SDK, complete with full audio/video controls and participant management.
  • • Developed AI-powered assistance for live agents, including automated response generation, conversation summarization, and contextual quick actions (e.g., calendar scheduling).
  • • Created a sophisticated and responsive agent inbox UI with conversation threading, toast notifications, and real-time status updates using modern frontend libraries.
Vurbalize App (SaaS Platform)
  • • Led the full-stack development of a multi-tenant SaaS platform using Next.js/React and FastAPI, featuring over 50 API endpoints for customer engagement workflows.
  • • Designed and implemented an enterprise-grade, multi-tenant Role-Based Access Control (RBAC) system with hierarchical permissions and JWT-based authentication.

Tech Stack: Python, FastAPI, Next.js, React, LangGraph, LangChain, Socket.IO, Celery, OpenAI GPT-4, Microsoft Azure, ChromaDB, Twilio Video SDK, Docker, Redis, MongoDB, HubSpot, Microsoft Graph API.

Pre-GenAI Phase

  • • Engineered a custom dual-encoder transformer architecture, incorporating a secondary encoder for spaCy-based NER to boost NLU accuracy by 25% and improve inference speed by 3x.
  • • Developed an end-to-end ML pipeline using PyTorch Lightning and Weights & Biases to train and optimize various BERT models, culminating in an incremental retraining system that reduced training time by 60%.
  • • Architected a scalable synthetic data generation system to produce over 100K training examples per run, enabling effective model training for new clients.

Tech Stack: PyTorch Lightning, Transformers, spaCy, Flask, MongoDB, Redis, GCP, Docker, Shopify API, Weights & Biases.

Hindustan Times

Data Science Intern (Dec 2020 – Mar 2021)
  • • Developed web crawlers using Beautiful Soup to collect and store news data in MongoDB.
  • • Implemented Python scripts to manage and upload data to an Elasticsearch cluster, visualized through Kibana.
  • • Applied BERTopic modeling to develop a timeline-based recommendation system for similar news articles.
  • • Designed and built a scalable knowledge graph for news articles using Neo4j and Cypher, leveraging NLP to extract and map entities and relationships.

Birla Science Center

Research Intern (May 2019 – July 2019)
  • • Contributed to the development of an automated surveillance system by applying computer vision algorithms for detection and tracking.
  • • Implemented and tested techniques like the TensorFlow Object Detection API and Kalman Filters to build a people counter and a warning system for restricted areas.

Projects

Tytona

AI Conversational Search Platform (Personal Project)
  • • Architected and developed a full-stack AI search platform from scratch, featuring a FastAPI backend, a responsive Next.js/React frontend with Zustand for state management, and a PostgreSQL database with Alembic for migrations.
  • • Built an advanced, agent-based AI system using LangGraph for multi-step reasoning, integrating multiple LLMs (OpenAI, local Ollama models) and search APIs (SerpAPI, Google) with streaming responses.
  • • Implemented a secure authentication system with Google OAuth2, JWTs, and secure cookie management, protecting over 15 RESTful API endpoints.
  • • Designed a scalable microservices-ready architecture and automated the deployment pipeline with custom bash scripts, Nginx reverse proxy, and systemd service management.

Full-Stack LeetCode Progress Tracker

Personal Project
  • • Developed a 3-tier application featuring a Chrome browser extension that communicates with a FastAPI backend and a Next.js/React frontend to track LeetCode progress in real-time.
  • • Built a comprehensive analytics dashboard with advanced data tables (TanStack) and animations (Framer Motion) to visualize user statistics and performance streaks.
  • • Implemented a spaced-repetition reminder system using automated email notifications and developed full CRUD operations for managing problems and solutions.
  • • Secured the application using Google OAuth2, JWT token management, and protected API routes.

Network on Chip Temperature Prediction

Research Project
  • • Utilized an open-source C++ simulator to generate thermal data for Network on Chip (NoC) architectures.
  • • Developed Python scripts using the subprocess library and Pandas to retrieve and process the simulated data.
  • • Employed AutoML frameworks like AutoGluon and AutoKeras to identify optimal Machine Learning and Deep Learning models for temperature prediction, visualizing results with Matplotlib.

Education & Certifications

BITS Pilani, Hyderabad Campus

August 2017 – May 2022

Hyderabad, India

M.Sc., Mathematics

B.E., Electronics and Communication

Certifications

Google Certified TensorFlow Developer

Mathematics for Machine Learning Specialization

DeepLearning.AI

Python and Statistics for Financial Analysis

Skills

AI & ML

LangGraphLangChainPyTorchTensorFlowTransformersRAGLLM OrchestrationMulti-Agent SystemsKnowledge GraphsComputer VisionNLPspaCyScikit-learnPandasWeights & Biases

Backend

PythonGoFastAPIFlaskSocket.IOSQLAlchemy

Frontend

ReactNext.jsTailwind CSSZustandTanstack QueryTypescript

Databases

PostgreSQLMongoDBRedisNeo4jChromaDBElasticsearch

Cloud & DevOps

Microsoft AzureGoogle Cloud Platform (GCP)AWSDockerNginxCI/CDGit