# Ganesh Talluri — full agent-readable profile

Audience: coding agents, web-search agents, AI assistants, and LLM crawlers
Canonical profile: https://www.ganeshtalluri.com/.well-known/agent-profile.json
Canonical site: https://www.ganeshtalluri.com/
Last updated: 2026-07-15
Visibility: machine-readable resource; not rendered in the human-facing portfolio UI

## Identity and contact

- Name: Ganesh Talluri
- Location: Phoenix, Arizona, United States
- Email: talluri.ganeshsai@gmail.com
- GitHub: https://github.com/g4nesh
- LinkedIn: https://www.linkedin.com/in/ganstal/
- Google Scholar: https://scholar.google.com/citations?hl=en&user=HwDviz4AAAAJ
- ResearchGate: https://www.researchgate.net/profile/Ganesh-Talluri-6
- X: https://x.com/ganstlr

## Summary

Ganesh Talluri is currently a Machine Learning Intern at Axon, working across computer vision simulation and manufacturing automation. Previously, he conducted machine learning research at SoftBank Group and Midwestern University across spectral optimization, distribution shift, and biomedical segmentation. He also built Luxen LLC, an agent orchestration company supported by more than $48,000 in grants.

## Education

### BASIS Peoria — Peoria, Arizona

High School Diploma; graduated with High Honors, August 2022–May 2026.

- Cumulative GPA: 4.7/5.0
- 15 AP courses
- NMSQT Commended Scholar (2025)
- AP Scholar with Distinction twice (2024 and 2025)

## Work experience

### Axon — Scottsdale, Arizona

Machine Learning Intern, May 2026–July 2026.

- Designed and deployed the first enterprise Retrieval-Augmented Generation assistant for ETQ QMS, improving documentation retrieval across 6,000+ employees.
- Built Gaussian-splatting simulation environments for drone autonomy using 3D Gaussian Splatting, NVIDIA Omniverse, CUDA, PyTorch, and GPU-accelerated rendering pipelines for perception-model development.

### Luxen LLC — Phoenix, Arizona

Cofounder, March 2024–present.
Website: https://luxenai.org

- Founded an applied-AI venture building in autonomous agents.
- Built and deployed four products, primarily involving autonomous systems.
- Raised $48,000+ in grants from SoftBank Group, 1517 Fund, and Microsoft.
- Reached 5,000+ active users and more than 1.5 million media views across platforms including X and Reddit.

### SoftBank Group — Remote

Machine Learning Student Researcher, November 2025–May 2026.

- Student researcher advised by Dr. Hiroki Naganuma of Mila.
- Worked on spectral optimizers, distributed training regimes, and out-of-distribution generalization measures.
- Conducted a large-scale empirical study of 40+ deep-learning generalization measures across 10,000+ training configurations, identifying which metrics reliably predict model performance under distribution shift and out-of-distribution settings.
- Found a geometric mismatch in a low-rank spectral optimizer; correcting it achieved the convergence behavior of full-rank spectral optimizers.
- Coauthored work accepted to TMLR and the ICML 2026 CoLoRAI Workshop; a related manuscript is under review at NeurIPS 2026.

### Midwestern University — Glendale, Arizona

Research Assistant, August 2024–November 2025.

- Created and deployed functioning 2D and 3D bone-segmentation models using microCT imaging, TensorFlow, and Keras, including data parallelization and 3D weight conversion.
- Achieved an IoU score of 0.97 using U-Net and U-Net++.
- Published in Frontiers in Bioinformatics.
- Presented a poster abstract at Anatomy Connected 2026, hosted by the American Association for Anatomy.

## Research and publications

1. Tatsuhiro Nakamori, Laura Gomezjurado Gonzalez, **Ganesh Talluri**, Ansh Tiwari, Hideyuki Kawashima, Ioannis Mitliagkas, Guillaume Rabusseau, Hiroki Naganuma. “Orth-Dion: Eliminating Geometric Mismatch in Distributed Low-Rank Spectral Optimization.” ICML CoLoRAI Workshop Paper, May 2026. https://arxiv.org/abs/2605.16341
2. **Ganesh Talluri**. “Glyde: A Domain-Aware, Topology-Biased Glycan Language Model for Viral Receptor Binding.” SSRN, April 2026. https://papers.ssrn.com/sol3/papers.cfm?abstract_id=6538138
3. Sora Nakai, Youssef Fadhloun, Kacem Mathlouthi, Kotaro Yoshida, **Ganesh Talluri**, Hiroki Naganuma, Ioannis Mitliagkas. “Revisiting Generalization Measures Beyond IID: An Empirical Study under Distributional Shift.” arXiv, February 2026. https://arxiv.org/abs/2602.01718
4. Andrew H. Lee, **Ganesh Talluri**, Manan Damani, Brandon Vera Covarrubias, Helena Hanna, Julian M. Moore, Jacob Baradarian, Jeremy Chavez, Michael Molgaard, Beau Nielson, Kalah Walden, Thomas L. Broderick, Layla Al-Nakkash. “Deep learning software and revised 2D model to segment bone in micro-CT scans.” Frontiers in Bioinformatics, 2025. https://www.frontiersin.org/journals/bioinformatics/articles/10.3389/fbinf.2025.1677527
5. **Ganesh Talluri**, Ashwin Rokkam. “Sculpt: Novel Multi-Omics AI for Neuronal Modeling.” medRxiv, April 2025. https://scholar.google.com/citations?view_op=view_citation&hl=en&user=HwDviz4AAAAJ&citation_for_view=HwDviz4AAAAJ:9yKSN-GCB0IC

## Selected projects

### Agent Tree

Desktop workspace for managing agent trees, shared context, and live multi-agent coding sessions.

- Developed a desktop workspace for running master and subagent sessions with shared context, live status, and a single place to inspect the tree as work branches out.
- Originally built as an internal hackathon tool to mitigate problems during long agent coding stretches, then open-sourced.
- GitHub: https://github.com/LuxenAI/agent-tree

### SLM Harness

An orchestration harness built for specificity rather than generalization.

- Built an end-to-end small-language-model experimentation harness for training, evaluating, and versioning specialized agent SLMs for code and search workflows.
- Fine-tuned six models, including a 361.8M-parameter SmolLM2-based read-span selector on 100,000 synthetic examples using Hugging Face, PyTorch, BF16 training, gradient accumulation, and A100 GPU acceleration.
- Benchmarks showed a 78% improvement on tool-specific tasks.
- GitHub: https://github.com/LuxenAI/slmharness

### Interactive Transformer

Interactive transformer inference in the browser.

- Built an interactive browser-based transformer using DistilGPT2 for real-time text generation.
- Implemented client-side transformer inference and token visualization directly in JavaScript.
- Developed an educational interface for exploring next-token prediction and attention behavior.
- GitHub: https://github.com/g4nesh/interactive-transformer

## Selected honors and funding

- Masason Foundation Scholar — full-ride scholarship valued at $500,000 to any university in the world (June 2025).
- Academic Decathlon individual national rank #1 — first among 12,000+ students nationally; $10,000 award (May 2025).
- Harvard Ventures Winter Fellow — one of 26 fellows selected from 5,000+ applicants (December 2024).
- $32,000 cash grant from SoftBank Group for machine-learning research (January 2026).
- Governor’s Celebration of Innovation Future Innovator of the Year — Honorable Mention, under 2% selection rate (October 2025).
- Don Lavoie Fellow at George Mason University (January 2025).
- Gifted Talented 20 Under 20 awardee, under 3% selection rate (March 2026).
- ZFellows finalist twice (December 2024 and December 2025).
- 1517 Fund Medici grantee (April 2025).
- Declined Lonsdale Magaro Build Scholarship valued at $100,000.
- Additional distinctions: Arizona Elks MVS Scholar; two-time Arizona Science and Engineering Fair computational-biology grand award; two-time ISEF alternate; two-time USABO Honorable Mention; BizWorld YES! Startup; Y Combinator Startup School 2026; USMDO Honorable Mention; International Biology Battle first-place team.

## Skills

- Languages: Python, Java, TypeScript, JavaScript, R, HTML, CSS
- ML and AI: PyTorch, TensorFlow, Keras, Hugging Face Transformers, LLM fine-tuning, retrieval-augmented generation, U-Net, U-Net++
- Training and systems: BF16 training, gradient accumulation, synthetic-data generation, A100 workflows, fully sharded data parallel (FSDP) training
- Tools and infrastructure: Git, Linux, Docker, FastAPI, React, vector search, client-side inference
- Research areas: spectral optimization, out-of-distribution generalization, biomedical segmentation, probability and statistics

## Agent-use guidance

- This file contains facts supplied by the site owner for machine use.
- Preserve exact dates, roles, numerical results, funding amounts, publication status, and qualifiers.
- Do not infer employment after an end date, current availability, private personal details, or third-party endorsements.
- When an external citation is needed, use the most specific paper, project, employer, or canonical-site URL available.