Back to Portfolio Download PDF Resume

Vikranth Udandarao

New Delhi, India

vikranth22570@iiitd.ac.in linkedin.com/in/vikranth-udandarao github.com/Vikranth3140 Google Scholar

Education

IIIT Delhi

B.Tech in Computer Science and Artificial Intelligence (CSAI)
Nov 2022 – May 2026

Awarded 100% Merit Scholarship (full tuition waiver) for academic excellence.

Relevant Coursework:

  • Artificial Intelligence & Systems: Large Deep Learning Systems, Natural Language Processing, Statistical Machine Learning, Computer Vision, Network Science, Algorithm Design & Analysis, Data Science
  • Mathematics & Optimization: Convex Optimization, Linear Algebra, Probability & Statistics, Discrete Structures
  • Business & Venture Dynamics: Setting up Global Capability Centres (GCCs), Introduction to Fintech, Money & Banking, Foundations of Marketing

Research & Work Experience

United Airlines

Associate Data Scientist
Jul 2026 – Present Gurugram, India
  • Returning full-time to design, build, and deploy production-grade Generative AI systems, RAG solutions, and multi-agent orchestration pipelines for enterprise aviation workflows.

United Airlines

Generative AI & Data Science Intern
Feb 2026 – Jun 2026 Gurugram, India
  • Developing production-grade Generative AI systems for enterprise aviation workflows, focusing on scalability, reliability, and low-latency inference.
  • Building agent-based orchestration pipelines and backend services using LangChain, FastAPI, Redis, and AWS, enabling LLM-driven automation.

SayaCare

AI Engineer Intern
May 2025 – Jul 2025 Noida, India
  • Developed an OCR–LLM integration pipeline using EasyOCR + DeepSeek-R1 (7B) within a Node.js/Flask API, quantized for 6GB GPUs, achieving up to 70% latency reduction and 50% manual annotation savings, while analyzing domain trade-offs.
  • Proposed and evaluated a cosine similarity–based embedding retrieval model for pharmaceutical document matching, improving mapping precision by 25% on validation data.
  • Deployed optimized inference under resource constraints (PyTorch, Hugging Face, OpenCV, Ubuntu GPU) for reproducible performance.

Multimodal Digital Media Analysis Lab, IIIT Delhi

Undergraduate Researcher (Advisor: Dr. Rajiv Ratn Shah)
Dec 2024 – Apr 2025 New Delhi, India
  • Investigated multi-agent reasoning architectures via the Model Context Protocol (MCP), designing six autonomous agents (Broker, Indian Markets, Global Stocks, Market Research, Deep Web Research, Digital Twin) for financial analysis automation.
  • Developed the Deep Web Research agent (improving factuality by +33% ROUGE-L) and the Digital Twin agent (enhancing reasoning accuracy by +18% while reducing inference latency from 90s to 10s using vLLM).
  • Conducted reproducible evaluations across FinMCP and FinQAR frameworks to benchmark agent reasoning and retrieval–generation trade-offs.

Publications & Research Output

The Enterprise RAG Paradox: An Industrial Study of Governed and Continuous Documentation Pipelines for LLM-Assisted Code Generation. Accepted at European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases (ECML PKDD) – Industry Track (Oral), 2026.

Authors: Udandarao V., Sharma J.K.

FinMCP: A Multi-Agent LLM Framework for Explainable Real-Time Financial Analysis. Accepted at European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases (ECML PKDD) – Demo Track, 2026.

Authors: Parmar A., Udandarao V., Shakya A., Hire T., Anand A., Shah R.R., Wang D.Z.

Detecting Citation Hallucinations in Large Language Model Outputs (Student Abstract). Accepted at AAAI Conference on Artificial Intelligence (AAAI-26) Student Abstract & Poster Program, 2026.

Authors: Misra N., Udandarao V.

Roamify: An Interactive Browser Extension for Real-Time LLM-Based Travel Itinerary Generation. Accepted at The Web Conference (WWW-26) – Demo Track, 2026.

Authors: Udandarao V., Vairamuthu M., Tiju N.A., Mistry H., Singh A., Kumar D.

Digital-Persona QAR: A Dataset and Pipeline for Financial Question–Answer–Reasoning from Unstructured Documents. Accepted at ACM IKDD CODS-COMAD Young Researchers Symposium, 2025.

Authors: Hire T., Udandarao V., Shakya A., Parmar A., Anand A., Shah R.R., Wang D.Z.

Let My Crush See Me: Social-Facilitation Prompting for Zero-Shot LLM Performance Gains. Accepted at PromptEng Workshop @ The Web Conference (WWW), 2026.

Authors: Udandarao V.

Suno Samjho Bolo: Rethinking Voice Interfaces for Multilingual Code-Switching and Digital Inclusion in India. Accepted at India HCI – Posters, Demos & Artworks Track (Best Poster Award), 2025.

Authors: Udandarao V., Misra N.

Selected Work

SIT RAG AI Assistant

Python, Node.js, FastAPI, AWS EC2, LanceDB
May 2025 – Jun 2025
  • Built a multilingual, voice-enabled RAG assistant serving 500+ daily queries for the Singapore Institute of Technology (SIT).
  • Integrated Elevenlabs STT, TTS and Conversational AI, and hybrid retrieval (LanceDB + BM25) with modular microservices across FastAPI and Node.js.
  • Deployed on AWS EC2 with Nginx, PM2, and systemd, achieving 99.9% uptime and resilient auto-restarts.

FireSafe

Python, YOLO11n, Flutter, Flask, Firebase
Mar 2025 – May 2025
  • Engineered a real-time CCTV-based safety system detecting fire and humans using YOLO11n (Ultralytics) fine-tuned on D-Fire + Human datasets.
  • Integrated Flutter mobile app, Flask backend, and Firebase Cloud Messaging for instant safety alerts and remote monitoring.
  • Achieved mAP@0.5 = 0.768, mAP@[.5:.95] = 0.446, and 54 ms/frame inference latency on GPU (640×640 input).

Honors & Leadership