Manav Pandey

Manav Pandey

Senior Machine Learning Engineer - Agentic AI

Scottsdale, AZ
AI Research
LLMs
Production ML
Agentic AI

Professional Focus

Architecting enterprise multi-agent AI frameworks serving thousands of users

Fine-tuning and optimizing large language models for production deployment

Leading AI safety research and discovering critical vulnerabilities in foundation models

Building scalable ML systems with comprehensive monitoring and security protocols

Programming Languages

🐍PythonExpert
🟨JavaScriptExpert
🔵GoProficient
JavaProficient
C++Proficient
🦀RustLearning

Employment Authorization

🇺🇸

US Citizen

About Me

In a Nutshell...

  • Senior ML Engineer at American Express, architecting enterprise-scale multi-agent AI frameworks.

  • Expert in using search algorithms (MCTS) and reinforcement learning to train and distill smaller, more efficient language models.

  • Passionate about proving that smart orchestration and tool-use can outperform massive foundation models in real-world applications.

  • Diverse experience from enterprise R&D to startups, focusing on production AI, model fine-tuning, and AI safety.

|

Experience

Senior Research Engineer - CTO R&D Team
American Express logo
American Express
Sep 2024 - Present

Leading enterprise-wide adoption of agentic AI systems and multi-agent frameworks. Primary driver of agentic AI strategy, presenting to 700+ attendees per session on best practices and implementation strategies.

Key Achievements:

  • Architected enterprise multi-agent AI framework using LangGraph and MCP with $35M ARR impact
  • Established company-wide standards for agentic AI: security protocols, evaluation metrics, tool-use patterns, and SLM post-training/distillation pipelines
  • Implemented meshnet adoption bringing traditional APIs to distributed agent communication systems

Technologies:

LangGraph
MCP
Small Language Models
Agentic AI
Multi-Agent Systems
Distributed Systems
Machine Learning Engineer - Research
Lightsource logo
Lightsource
Feb 2024 - Sep 2024

Specialized in fine-tuning large language models and optimizing inference pipelines for production deployment. Achieved significant improvements in model performance and user satisfaction through advanced training techniques.

Key Achievements:

  • Fine-tuned Mistral 7B and Mixtral 8x7B models achieving 94% user satisfaction using PPO reinforcement learning and supervised fine-tuning, compared to initial 65%
  • Optimized inference pipelines with INT4-FP8 quantization techniques reducing latency by 60% across heterogeneous GPU hardware
  • Implemented spherical interpolation model merging improving multilingual performance by 23% while maintaining base model capabilities
  • Deployed production-ready transformers serving 100K+ daily queries with comprehensive monitoring and A/B testing infrastructure

Technologies:

Mistral
Mixtral
PPO
RLHF
Model Quantization
GPU Optimization
A/B Testing
Production ML
Director of AI
Curiouser logo
Curiouser
Dec 2023 - Sep 2024

Led technical direction for engineering team developing high-EQ conversational AI platform. Focused on cognitive-inspired agent architecture and emotional intelligence simulation.

Key Achievements:

  • Led technical direction for 5-person engineering team, developing high-EQ conversational AI platform using multi-LLM architecture
  • Engineered cognitive-inspired agent architecture for behavioral analysis and emotional intelligence simulation, with autonomous knowledge graph creation
  • Employed use of task queues with celery and graph-based agentic systems to create parallel, real-time conversational processing

Technologies:

Multi-LLM Architecture
Knowledge Graphs
Celery
Graph-based Systems
Conversational AI
Emotional Intelligence
Software Engineer
American Express logo
American Express
Aug 2022 - Feb 2024

Built production conversational AI systems and implemented comprehensive monitoring across microservices architecture. Focused on scalable AI deployment and system observability.

Key Achievements:

  • Built production conversational AI using BERT models, achieving 90% accuracy across 1M+ monthly interactions
  • Implemented comprehensive APM monitoring across 100+ microservices using Splunk and OpenTelemetry

Technologies:

BERT
Conversational AI
Microservices
Splunk
OpenTelemetry
APM Monitoring
Machine Learning Research Assistant
Texas A&M University logo
Texas A&M University
Oct 2021 - May 2022

Conducted research on sentiment classification and cross-border threat detection using advanced machine learning techniques. Applied NLP and deep learning to process large-scale social media and news data.

Key Achievements:

  • Enhanced SVM model accuracy from 79% to 94% through kernel optimization for sentiment classification and cross-border threat detection
  • Applied NLP and deep learning techniques to aggregate and process news/social media data for US-Mexico-Canada supply chain assessment

Technologies:

SVM
Kernel Optimization
NLP
Deep Learning
Sentiment Analysis
Social Media Analytics

Education & Certifications

Academic background and professional certifications in AI and machine learning

Education
Georgia Institute of Technology logo

Masters in Computer Science (Machine Learning)

Georgia Institute of Technology

2025 - Present

Texas A&M University logo

Bachelors in Computing

Texas A&M University

2018 - 2022

Certifications
Anthropic logo

Anthropic Bug Bounty Program

Anthropic

2024

Coursera logo

Coursera AI Instructor

Coursera

2023

Core Competencies
Foundation Models
Reinforcement Learning
Production AI Systems
Multi-Agent Systems
AI Safety Research
LLM Fine-tuning
Model Optimization
Agentic AI

Technical Skills

LLMs & Deep Learning

PyTorch
TensorFlow
Transformers
RLHF/PPO
Supervised Fine-tuning
Distillation
DPO
LangChain
LlamaIndex

AI/ML Engineering

vLLM
TensorRT
GGUF/AWQ/EXL2 Quantization
Distributed Training
CUDA
Model Merging
RAG Architectures

Software Engineering

Python
C++
TypeScript
FastAPI
Docker
Kubernetes
AWS SageMaker
Multi-agent Systems
MCP
Production ML

Research & Safety

AI Safety Research
Adversarial Testing
Model Validation
Bug Bounty Programs
Security Protocols
Evaluation Metrics

© 2025 Manav Pandey. All rights reserved.

0%