Dev Seth

cto · xantyr · ai architect

Six years building AI that enterprises actually trust.

I've fine-tuned models for individuals and Fortune 500s, shipped production AI across pharma, retail, real estate, and government — and I'm now building the platform that lets any business own its model the same way.

0 yrs

in AI & ML

0+

AI engineers led (grown from a team of 5)

0+

products shipped from idea to paying customer

0%

accuracy, most accurate US housing forecast of 2023

I'm the co-founder — I write the code

now building

Vibe code your own private & affordable LLM.

Xantyr turns a company's own data into a specialised model it owns outright — structure it, fine-tune it, deploy it. No ML team required.

Visit xantyr.com ▸

the platform.

xantyr.

the trajectory

From a team of 5 to leading 25+.

Jan 2021Oct 2024Nov 2025
5Sr. Data Scientist, Home.LLC

selected work

Case studies that moved the needle.

Defence

Sovereign AI for national defenceconfidential

Currently building secure, air-gapped AI systems with the Indian government for the Army and Navy — predictive maintenance, threat detection, sovereign models trained without any data leaving controlled infrastructure.

Real estate

Conversational AI agents: Sasha & Selina

Architected two production AI agents — Sasha for real-time sales objection handling, Selina with hyper-local insights across 33,000+ US neighborhoods. Scaled to $50K+ MRR.

Pharma

Life-critical supply chain optimisation

Built an algorithmic system for a UK pharma manufacturer producing last-stage cancer injections with a 3-day half-life, where delivery delays are life-threatening.

Retail

AI-powered inventory management at scale

Replaced a 10-person manual team with an AI agent for a UK retailer operating 22+ stores and 5 warehouses.

Government

24/7 government service automation

Deployed a voice AI agent for a government entity in Oman, turning 12-hour manual operations into round-the-clock automated service.

Pharma

Pharma master data management

Automated golden record creation and entity resolution across 5 to 7 disparate datasets for a pharmaceutical company.

technical stack

Deep across the full AI stack.

LLM & agentic AI

Retrieval-augmented generationLangChain & LangGraphLLM fine-tuning (LoRA/QLoRA)Vector DBs (Pinecone, Chroma)TTS & STTPrompt engineering

Cloud & MLOps

AWS SageMaker & LambdaDocker & KubernetesCI/CD pipelinesFastAPI & FlaskMLflowTerraform

Data science & ML

Python & SQLPyTorch & TensorFlowScikit-learnTableau & Power BIApache SparkA/B testing

Leadership

System architectureTechnical leadershipAgile / ScrumStakeholder managementP&L ownershipCross-functional teams

Currently building with

Claude CodeClaude (Design)GemmaNext.js + ReactNeonStripeModal + Unsloth (LoRA/QLoRA)VercelCloudflare R2Fly.io

off the resume

Things that don't fit in a bullet point.

Uber kept surging near my college, so I started saving drivers' numbers after every ride. Eventually I had a WhatsApp group of 40-something drivers, plus my friends. Need a ride? Text the group. No surge, no app fee, and drivers liked it because they got a guaranteed ride when free.

Building Xantyr. Always happy to talk shop.

Ready to connect — always up for a conversation about AI, defence-grade infrastructure, or anything hard and technical.