ABHIMANYU KUMAR

Dronacharya College of Engineering

Bachelor of Technology

May 1, 2014 – July 1, 2018

Modern Era Convent

All India Senior School Certificate Examination

May 1, 2012 – March 1, 2014

Modern Era Convent

All India Secondary School Examination

May 1, 2010 – March 1, 2012

Joy IT Solutions Pvt Ltd

Senior Development Engineer

December 1, 2025 – Present

Avizva Solution Private Limited

Senior Data Engineer

September 1, 2021 – March 1, 2025

OYO Rooms and Hotels

Product Lead

November 1, 2020 – September 1, 2021

Droom Technologies

Product Analyst

June 1, 2019 – September 1, 2020

Fareportal Private Limited

Machine Learning Executive

August 1, 2018 – May 1, 2019

Agentic AI Chatbot for Internal Marketing Operations

April 29, 2026 – Present

Objective: To build an intelligent, multi-agent AI chatbot that empowers internal marketing teams to independently resolve queries, manage support tickets, and report bugs — reducing dependency on manual processes and improving operational efficiency through conversational AI. Key Highlights: Designed and developed a multi-agent AI system using Google ADK and Vertex AI, with a root orchestrator agent routing tasks to specialized sub-agents based on user intent. Built a knowledge retrieval sub-agent leveraging Vertex AI Search to fetch contextual answers from internal websites and historical Zendesk tickets, enabling accurate self-serve query resolution. Developed a Zendesk integration sub-agent supporting end-to-end ticket lifecycle management - creation, updation, and status tracking - through a natural language conversational interface. Integrated Google's internal bug tracking tool via a dedicated sub-agent, allowing users to log and track bugs directly through the chatbot without switching tools. Designed and executed test cases for individual sub-agent functionality and end-to-end multi-agent orchestration flows, validating routing logic, tool invocation, and response accuracy. Performed regression testing post-updates to ensure reliability and consistency across all sub-agents and their external API integrations.

Vehicle Listing Automation Framework

April 29, 2026 – Present

Objective: Designed and implemented a robust automation framework to synchronize vehicle listings from Orange Book Value (OBV) — a pricing and listing tool owned by Droom — to Droom's main marketplace platform and other pre-approved third-party portals. Description: Built a scheduled data ingestion pipeline using cron jobs that ran twice daily to fetch new, user-approved listings from the OBV MongoDB database. Automated the listing process using Selenium to simulate user interaction and fill listing forms on Droom's web interface. Created dynamic field mapping modules to adapt OBV data structure to Droom's frontend form requirements. Implemented logging and error handling with screenshot capture for failed submissions, ensuring easy traceability and retries. Maintained data consistency across platforms without using backend APIs, complying with security and integration protocols. Reduced manual listing overhead by 90%, enabling near real-time synchronization of seller listings across Droom platforms. Improved operational efficiency and listing accuracy, contributing to a better seller experience and higher inventory freshness.

Multi-Agent Healthcare Support Chatbot using LLM & RAG

April 29, 2026 – Present

Objective: Developed an intelligent, real-time AI assistant for healthcare customer support teams, streamlining both IVR and chat-based interactions using LLMs, RAG, and speech-to-text technologies. Description: Designed and deployed a multiagent, multichannel AI chatbot system to handle real-time conversations for both chat and IVR calls within the healthcare domain. Built backend services using FastAPI to manage session lifecycles, maintain call states, and orchestrate intelligent routing of user queries across multiple agents. Implemented audio stream handling using WebRTC, converting audio to text using a self-hosted Whisper model and third-party services like Deepgram, depending on channel and latency requirements. Integrated a dynamic RAG pipeline for intelligent document retrieval, where orchestrator agents evaluated the query type and context, then routed it to specialized responder agents to generate accurate responses. Applied validation layers before delivering responses to users, ensuring compliance with healthcare standards and internal policy. Developed real-time agent assist tools for customer support staff, including: Live transcription with entity tagging (e.g., Claim ID, Plan ID) to extract actionable data. One-click navigation triggers from transcript to relevant internal system pages. Smart assistant features like auto-suggested replies, conversation summarization, task reminders, and note-taking. Enabled seamless transition from bot to human support agent when users requested escalation, preserving conversation context.

AI Testing Framework for Healthcare Chatbot Evaluation

April 29, 2026 – Present

Objective: Built an automated testing framework to evaluate the performance, accuracy, and contextual reliability of a healthcare-focused LLM-based chatbot. Description: Developed a robust AI testing framework to validate chatbot responses against pre-defined test cases curated by QA and clinical teams. Automated the ingestion of test cases from sources like Confluence, including input queries and expected ground truth responses. Dynamically generated test case variants across dimensions such as complexity (simple/multi-context), user demographics, and domain-specific factors, to simulate diverse real-world scenarios. Executed test cases against the production bot API and captured system-generated responses. Implemented an evaluation engine to compare actual vs expected results using multiple AI quality metrics: Answer Relevance, Context Recall & Precision, Faithfulness to Source, Latency (Response Time). Produced structured test reports with pass/fail classifications to report portal, highlighting failure types for debugging and model fine-tuning.

JIRA & Confluence Project Setup Automation Tool

April 29, 2026 – Present

Objective: Streamlined the setup of standardised JIRA projects and Confluence spaces for product and engineering teams using a reusable automation framework. Description: Developed a backend service using Python, Django, and the Django REST Framework to automate the creation of JIRA and Confluence instances from predefined templates. Designed a class-based schema in MongoDB to store and manage reusable project templates, hierarchies, and configuration settings. Integrated with Atlassian APIs to dynamically generate: JIRA Projects with custom issue types, workflows, boards, and permissions. Confluence Spaces with standard page layouts, documentation trees, and macros. Enabled teams to initiate full project environments with a single API request, reducing manual setup time from hours to minutes. Included logging, role-based access, and validation layers to ensure consistency across different teams and departments.

Auto-Mapper: LLM-Based CSV Attribute Mapping System

April 29, 2026 – Present

Objective: Automated the process of mapping user-uploaded CSV attributes to standardized portal attributes using natural language understanding and LLMs. Description: Developed a backend pipeline to ingest and analyze CSV files uploaded by users for system integration and data onboarding. Used LLMs to semantically understand column headers and values, mapping them to internal standardized schema attributes (e.g., user ID→ member_id, contact no. → phone_number). Applied Pydantic parsers to validate and reformat extracted data according to schema requirements. Implemented a rules-based fallback for unmapped or ambiguous attributes using synonym libraries and contextual matching. Built the pipeline using Python and integrated with portal systems to enable near real-time onboarding of structured data. Reduced manual mapping time significantly and improved onboarding accuracy.

PDF-to-JSON Converter using RAG & Self-Hosted LLMs

April 29, 2026 – Present

Objective: Automated the transformation of unstructured PDF documents into structured JSON using Retrieval-Augmented Generation (RAG) and fine-tuned LLMs. Description: Developed a document processing pipeline that retrieves PDFs from AWS S3, applies RAG-based prompting, and extracts structured data into JSON format. Used self-hosted, fine-tuned LLMs to extract contextually accurate information, even from semi-structured documents like forms and benefit summaries. Implemented OCR fallback for scanned documents and integrated custom validation logic for mandatory field checks. Enabled scalable ingestion of healthcare and policy documents, feeding downstream systems with clean, machine-readable data.

Hotel Recommendation Engine using Cosine Similarity

April 29, 2026 – Present

Objective: Recommended alternative hotels to users based on similarity and geographic proximity to enhance booking conversions and user experience. Description: Built a content-based recommendation system to suggest hotels that are most similar to a user's current selection and located within a specified radius. Engineered feature vectors using attributes such as hotel type, amenities, price range, star rating, and user reviews. Applied cosine similarity to calculate semantic similarity between hotels and return the top matches within a defined geographic vicinity. Designed the system to power upsell suggestions and fallback recommendations in case of sold-out listings. Delivered as a scalable microservice integrated with the hotel search and booking pipeline.