ta.Tejas Arya
Let’s talk

Applied Scientist at Tekion

AI that understands.
Systems that deliver.

I’m Tejas. I turn complex customer questions into useful answers—with agents, intelligent search, and production machine learning.

Explore my work Résumé

M.Tech, AI & ML · IIT Hyderabad Based in Bengaluru

My focus01 / Vehicle intelligence
Customer question

“A Toyota SUV under $40K
with premium wheels.”

Understand. Find. Verify.

Vehicle Enquiry + Inventory Expert

A useful, grounded answer

Vehicle matches, equipment, and availability.

See how it works

Production AI.
Measurable impact.

10,000+call transcripts / dayCall intent classification
5,000+AI-assisted emails / dayCRM Pro
$6Mattributable annual revenueCRM Pro

Selected work

Built for the real world.

From vehicle conversations
to healthcare intelligence.

TEKION CORP02

Context-aware email assistance

CRM Pro / Generative AI email assistance

Personalized email suggestions grounded in CRM history and conversations—serving 30% of dealership email volume and driving $6M in attributable annual revenue.

Generative AIKafkaRedis
How it works

Context-aware topic ranking and summaries of calls, emails, and chats build high-signal prompts for personalized, compliance-aligned drafts.

Asynchronous Kafka events, versioned MongoDB context, Redis caching, and I/O profiling support low-latency, zero-downtime generation for 5,000+ emails per day.

CARELON GLOBAL SOLUTIONS03

Healthcare document intelligence

Healthcare / Document intelligence & retrieval

Machine learning and NLP pipelines that transform handwritten records and complex medical documents into searchable, usable information.

PyTorchTrOCRLayoutLMv3RAG
How it works

TrOCR delivered approximately 95% handwritten recognition accuracy. LayoutLMv3 improved NER performance by 20%, and fine-tuned BERT achieved approximately 85% document QA accuracy.

A LangChain RAG system processed 10,000+ medical documents daily. Vector search cut query latency by 40%; an OpenCV pipeline handled 500+ documents per hour.

IIT HYDERABAD · COURSE PROJECT04

Clarification question generation

Clarification Question Generation / 2022–2023

Fine-tuned BART on Amazon QuAC to generate useful follow-up questions, with a schema-mapping module that identifies missing context.

BARTFine-TuningSequence-to-Sequence
Explore the repository

Experience

Research to production.

Building useful AI across automotive retail and healthcare.

The full résumé
JUL 2024 — PRESENT CURRENTTekion CorpApplied Scientist I · Bengaluru

Building production-grade agentic and generative AI for automotive retail.

  • Developed vehicle-domain capabilities within a multi-agent conversational commerce platform, working with LangGraph and tool-based retrieval.
  • Worked on Vehicle Enquiry and Inventory Expert workflows: understanding follow-up questions, finding relevant vehicles, and checking live availability.
  • Built inventory search with structured filters, semantic matching, and controls for pricing, test-drive eligibility, and tool failures.
  • Scaled call intent classification to 10,000+ transcripts daily with Docker, Kafka, and automated backpressure handling.
  • Built AI email assistance powering 5,000+ emails daily and $6M in attributable annual revenue.
  • Established observability and MLOps with OpenTelemetry, MLflow, and Databricks.
Agentic AIGenerative AIMLOps
AUG 2023 — JUL 2024Carelon Global SolutionsAssociate Software Engineer III · Hyderabad

Developed machine learning and NLP systems for healthcare analytics.

  • Built handwriting recognition, named entity recognition, and document QA pipelines with TrOCR, LayoutLMv3, and BERT.
  • Deployed retrieval-augmented QA for 10,000+ complex medical documents daily.
  • Reduced semantic search latency by 40% through vector database integration.
  • Paired OpenCV document processing with interactive Streamlit dashboards.
NLPComputer VisionRAG

Research & education

A foundation in curiosity.

IEEE PUBLICATION · 2024

Vehicle Speed Prediction
using Deep Learning

A CNN–GRU hybrid architecture, extended with a Transformer forecaster, that combines wheel-speed, acceleration, and steering-angle signals for temporal prediction.

Hyundai-sponsored M.Tech thesis, with model interfaces aligned to ABS hardware integration constraints.

Read the publication
EDUCATION
2021 — 2023

Indian Institute of Technology
Hyderabad

M.Tech · Artificial Intelligence & Machine Learning

2017 — 2021

Netaji Subhas Institute
of Technology, Delhi

Bachelor of Engineering

Toolkit

From model to production.

01 / Agents & retrieval

Understand and act

LangGraph · LangChain · ReAct · MCP · RAG / GraphRAG · Elasticsearch · Vector search

02 / Machine learning

Train and evaluate

PyTorch · TensorFlow · Transformers · BERT / BART · Fine-tuning · NLP · Computer vision

03 / Production systems

Build for scale

Python · AsyncIO · Docker · Kafka · Redis · MongoDB · MLflow · OpenTelemetry · AWS / Azure

Let’s connect

Have something
worth building?

Let’s talk about agents, applied AI, or your next hard problem.

tejas23arya@gmail.com Find me on GitHub Download my résumé