Veer Khot | AI Architect

Hi, I'm Veer Khot

AI Architect with 8+ years of experience delivering enterprise-scale AI systems across BFSI, healthcare, and climate-tech domains.

End-to-end expertise - from advanced modeling and reinforcement learning to PEFT-based LLM fine-tuning (LoRA, QLoRA), multi-reasoning agentic AI systems, and production-grade RAG architectures deployed on GCP & AWS.

Technical Skills

GenAI & Agentic Systems

End-to-End RAG Architectures
Multi-Agent & Multi-Reasoning AI
LLM Fine-Tuning (PEFT, LoRA, QLoRA)
RLHF & Reinforcement Learning
Knowledge Graph Integration
LLM Evaluation & Guardrails

Modeling & ML Engineering

Ensemble Models (XGBoost, RF, GBM)
Logistic & Linear Regression
Deep Learning (PyTorch, Autoencoders)
NLP & Transformer Models (BERT)
Time Series Forecasting (ARIMA)
Knowledge Distillation & Optimization

Data Science & Analytics

Statistical Analysis & Hypothesis Testing
Feature Engineering
Experiment Design (A/B Testing)
Bayesian Modeling & MMM
Model Evaluation & Validation
Business Impact Modeling

Cloud & Production Systems

GCP & AWS
Vertex AI & SageMaker
ETL & Scalable Data Pipelines
PySpark & Distributed Training
Vector Databases (FAISS, chromaDB)
MLOps & Model Monitoring

Work Experience

Citi — Data Scientist

Oct 2023 – Present | Bangalore, India

AI-Driven Regulatory Control Automation

Built multi-agent GenAI system using Vertex AI (Gemini) and FAISS to extract regulatory controls across 29 products.
Designed end-to-end pipeline from document ingestion to automated rule codification in PySpark.
Deployed 20+ production control scenarios.
Mitigated multi-million-dollar risk exposure.

Behavioral Segmentation & Explainable AI

Segmented card portfolio and diagnosed YoY revenue shifts.
Applied SHAP-based explainability to identify revenue drivers.
Integrated LLM-based executive summaries for leadership insights.

IEB Predictive Modeling

Developed ensemble models (XGBoost, RF, GBM).
Enabled potential $50M+ impact in portfolio protection.

Philips Healthcare — Data Science Manager

Apr 2022 – Sep 2023 | Bangalore, India

Clinical NLP Intent Recognition

Built transformer-based system using Clinical BERT.
Designed end-to-end ML pipeline and deployment workflow.
Improved conversion by ~20%.

Bayesian Market Mix Modeling

Architected Bayesian MMM using PyStan.
Delivered executive-level ROI optimization insights.

Ambee — Data Scientist II

Aug 2019 – Apr 2022 | Bangalore, India

Climate Forecasting & MLOps Pipeline

Built scalable ETL + ML pipelines on AWS (Glue, SageMaker).
Developed XGBoost & ARIMA forecasting models.
Implemented automated MLOps workflows for deployment.

UpYourGame — Data Scientist

Aug 2018 – Jul 2019 | Bangalore, India

Automated ML Pipeline

Designed AutoML-driven architecture for predictive analytics.
Reduced retraining time by ~90%.

Portfolio

Multi-Agent Regulatory Control Automation

Vertex AI Gemini FAISS PySpark Multi-Agent AI

Extracted regulatory controls from unstructured legal documents.
Designed end-to-end RAG + rule codification pipeline.
Automated compliance control scenarios in production.

Proactive AI for compliance & risk mitigation.

Enterprise Project

Agentic Customer Support AI System

RAG Multi-Reasoning LLM Orchestration Escalation Logic

Built multi-stage reasoning workflow for issue resolution.
Integrated retrieval + contextual generation.
Designed escalation mechanism for sensitive cases.

Scalable agentic AI for customer operations automation.

View on GitHub

Custom GPT Model (PyTorch)

Transformers PyTorch Tokenizer LLM Training

Built GPT architecture from scratch.
Implemented tokenizer & training loop.
Demonstrated transformer fundamentals.

Deep understanding of LLM internals & training workflows.

View on GitHub

LoRA Fine-Tuning + RAG Integration

PEFT LoRA QLoRA HuggingFace

Fine-tuned GPT-2 using parameter-efficient techniques.
Integrated retrieval augmentation.
Reduced compute cost while maintaining performance.

Efficient LLM adaptation for enterprise deployment.

View on GitHub

Articles

Why Human-in-the-Loop Architecture is the Creative Core of Agentic AI Systems

March 2026 · 8 min read

Hallucination is structural in LLM systems. This article explores how human-in-the-loop design becomes the most creative and resilient layer in production-grade agentic AI architectures.

Read Article →

Why Hallucination Is a System-Level Problem, Not a Model-Level Problem

---------- . ----------

When it comes to highly regulated domains, hallucination is not just a model limitation, it becomes a system design challenge. The issue is not that LLMs make mistakes; it is that the overall architecture must anticipate and handle those mistakes responsibly.

Coming soon

Let’s Collaborate

Building with LLMs but struggling to make them reliable in production?

I help move from experimentation to production-grade AI — designing systems that are measurable, controllable, and built for real-world constraints.

If you’re serious about deploying GenAI the right way — with proper architecture, evaluation, and human oversight — let’s talk.