Hi, I'm

Sanjay Nadhavajhala

Head of AI @ F2 · Full-stack AI builder — infra to agents.

Building the best finance agent in the world · 3 patents · 1M+ model downloads
Get in touch

About

I've spent the last decade building AI systems that ship — from speech models at IBM Watson to the world's first open-source AIOps platform at SUSE, to open-source tool-calling language models downloaded over a million times, to the #1 ranked spreadsheet agent in the world at F2.

I'm a full-stack AI builder. I design cloud infrastructure, architect agent runtimes, train and fine-tune models, and ship products end-to-end. I was the first AI hire at Rancher Labs (acquired by SUSE, which IPO'd at ~€5B), and today I'm a founding engineer and a technical lead at F2, where I set the product vision and built the systems that took us from zero to multi-million dollar ARR.

I've also served as a judge for the Bezos Earth Fund's $100M AI for Climate and Nature Grand Challenge and hold three U.S. patents spanning AI security, IoT privacy, and 3D avatar generation.

I like to understand the full stack — from infrastructure to product.

Experience

Sep 2025 – Present

Head of AI · Founding Engineer · F2 F2

Building the best finance agent in the world, 0 → multi-million ARR

A founding engineer and a technical lead at F2, where I set the product vision and architecture for agentic AI in finance — driving the company from zero to multi-million dollar ARR. Designed and built the core systems that define F2's product: cloud infrastructure on AWS and GCP, the agent runtime (plan mode, durable workflows for long-running autonomous agents), and MCP servers for Excel and PowerPoint. Built a high-throughput document-understanding pipeline that processes millions of financial documents per hour — extracting structured data, tables, and figures from PDFs at scale. Architected a real-time shared filesystem on EFS with a custom synchronization layer, giving users and agents a Google Drive-like live collaboration experience where both operate on the same workspace simultaneously. Engineered a hybrid-search architecture combining institutional knowledge with live market data to surface opportunities and validate assumptions faster. Our Excel Agent ranks #1 in the world — scoring 95.25% on SpreadsheetBench Verified, the standard public benchmark for evaluating spreadsheet agents (NeurIPS 2024).
f2.ai
Mar 2025 – Sep 2025

Head of AI · Arc Arc

Spun out F2 as independent company

Led AI strategy and development at Arc before spinning out F2 as an independent company. Defined the product vision and technical direction for AI-powered financial tools that became the foundation of F2.
joinarc.com
Jul 2023 – Mar 2025

Lead AI Engineer · Acorn Labs (now Obot AI) Acorn Labs (now Obot AI)

Rubra: 1M+ downloads, novel block expansion training

Led AI engineering for Acorn Labs, driving development across the Obot agentic platform and the Rubra suite of open-source language models. Rubra was one of the earliest collections of models fine-tuned specifically for tool calling — released before "agentic AI" entered the mainstream lexicon. The models surpassed 100K downloads in the first week and have been downloaded over 1 million times to date. Developed a novel post-training approach using block expansion — adding new transformer blocks and training only the expanded parameters to teach tool-calling behavior while avoiding catastrophic forgetting of the base model's capabilities. This technique remains largely underexplored in the field. Defined model strategy, built infrastructure for fine-tuning and evaluation, and shipped production-ready AI capabilities across the platform.
huggingface.co
Feb 2021 – Jul 2023

Engineering Manager · SUSE SUSE

World's first open-source AIOps (Opni), SUSECON keynote

Promoted to Engineering Manager following SUSE's acquisition of Rancher Labs (backed by EQT, which took SUSE public on the Frankfurt Stock Exchange in May 2021 at a ~€5B valuation). Created and led three product lines: Opni, Rancher Prime AI Assistant, and Rancher Observability. Opni was the world's first open-source AIOps platform — competing directly with Datadog — and transformed Rancher's observability story from a basic Prometheus operator into an intelligent, AI-driven offering with log anomaly detection, automated alerting, and multi-cluster insights. Rancher Prime AI Assistant was the first Kubernetes AI assistant with generative capabilities. Owned end-to-end delivery across all three products — software architecture, AI model design, product management, and hiring. Managed a globally distributed team of 10 Kubernetes and AI engineers. Keynote speaker at SUSECON 2023 in Munich, presenting Rancher Prime AI and AI Insights to the global SUSE community.
opni.io
Nov 2020 – Feb 2021

Staff Software Engineer · Rancher Labs → SUSE Rancher Labs → SUSE

First AI hire, acquired by SUSE/EQT

Joined Rancher Labs as its first AI hire to build the company's AI and machine learning capabilities from the ground up. Within weeks, Rancher Labs was acquired by SUSE in a deal led by EQT. Designed the foundational architecture for what would become Opni and the Rancher AI Assistant — both firsts in the Kubernetes ecosystem.
Sep 2018 – Oct 2020

ML Engineer, Watson AI · IBM IBM

Multi-GPU speech training, AIOps, 3 patents

Speech to Text — Introduced multi-GPU training to the speech pipeline. Developed a novel memory-mapped data loader that dramatically increased effective RAM utilization, yielding 38% faster model training. Designed a balanced-load multi-GPU training strategy that improved training speed by 61% and reduced Word Error Rate by 1%. Built and shipped production broadband and narrowband STT models for Spanish and Brazilian Portuguese. AIOps — Designed the log anomaly detection model architecture and led a team that built a full simulation environment — cloud-native applications on Kubernetes/OpenShift with LogDNA, Humio, Sysdig, Prometheus, PagerDuty, and ServiceNow — to replicate production client environments and validate models through fault injection. Customer Care Analytics — Developed novel evaluation metrics (Metric Error Rate, Sentiment Error Rate) for assessing composite ML systems with a speech-to-text component.
May 2017 – Aug 2017

ML Engineer, Extreme Blue · IBM IBM

Veracity project for Watson Discovery

Selected for IBM's Extreme Blue program — a highly selective technical leadership internship where small teams build and present product prototypes to IBM executives. Led development of the Veracity project, a machine learning system for identifying questionable and unreliable content within IBM Watson Discovery. Presented the project at Expo Day to senior IBM leadership.
Jan 2017 – May 2017

Data Scientist · Disney Disney

Box office projections, Super Bowl ad placement

Graduate intern on the Management Science team. Built theatre-level box office projection models using multi-layer perceptrons and developed the data layer on Apache Spark to process and analyze millions of data points across Disney's content portfolio. Work directly influenced Disney's marketing decisions, including determining Super Bowl ad slot placements for Disney and Pixar films during the 2017 Super Bowl.
Jun 2015 – Sep 2015

Software Engineer · Amazon Amazon

Built 1-click checkout → became "Buy Now" button

Intern on the Amazon B2B team. Developed the 1-click checkout experience, which shipped to Amazon B2B and eventually became the "Buy Now" button on Amazon.com.

Projects

F2 F2

Agents for finance

The AI platform for private markets investors. F2 automates data room processing, generates IC-ready reports, and combines institutional knowledge with live market data. I built LLMExcel — an LLM-agnostic tool that gives any model the ability to manipulate Excel files, ranked #1 in the world on SpreadsheetBench Verified (95.25%, NeurIPS 2024). As LLMs converge in raw intelligence, the edge is in tools and orchestration — that's what I build.

Rubra Rubra

1M+ downloads

Acorn Labs (now Obot AI)

One of the earliest open-source collections of language models fine-tuned specifically for tool calling — enabling LLMs to use tools, query APIs, and execute multi-step workflows out of the box. Built using a novel block expansion post-training method that preserves base model capabilities while teaching new behaviors.

Opni Opni

SUSE

The world's first open-source AIOps platform — providing intelligent log anomaly detection, automated alerting, and multi-cluster observability across Kubernetes environments. Competed directly with Datadog.

AI for Climate Grand Challenge — Judge AI for Climate Grand Challenge — Judge

Bezos Earth Fund

Served on the expert review panel for the Bezos Earth Fund's $100M AI for Climate and Nature Grand Challenge — evaluating grant proposals at the intersection of modern AI, climate science, and biodiversity conservation.

Speaking Engagements

FinTech Forward Summit

Stanford GSB · Stanford, CA

2026

Fine-Tuning LLMs for Multi-Turn Function Calling

MLOps Community · San Francisco, CA

2024

Keynote Speaker — AI for Multi-Cluster Kubernetes

SUSECON 2023 · Munich, Germany

2023

Multi Cluster Observability and AIOps with Opni

Cloud Native Rejekts NA · Detroit, MI

2022

Setting up a Logging Stack in 5 Minutes

Cloud Native Rejekts · Detroit, MI

2022

Opni

SUSECON 2022 · Remote

2022

Let's Learn Opni

Kubesimplify · Remote

2021

Patents

Artificially Intelligent Security Incident and Event Management

US-11665180-B2 · 2023

AI-driven SIEM system using attention-based deep neural networks and transfer learning for real-time classification of malicious activity.

Managing Anonymous Network Connections

US-11265305-B2 · 2022

Hierarchical anonymous credential system for IoT devices using PKI and Zero-Knowledge Proofs.

Creation of Photorealistic 3D Avatars Using Deep Neural Networks

US-11138781-B1 · 2021

GAN-based pipeline for generating photorealistic 3D avatars from 2D images using style transfer and meta-learning.

Education

University of Southern California

M.S. Computer Science

Machine Learning

University of Southern California

UC San Diego

B.S. Computer Science

UC San Diego

Get in touch

Want to collaborate or just say hi? Reach out — I read every message.

[email protected]