Full LLM lifecycle — data, pre-training, post-training (SFT + RL), evaluation, and large-scale deployment.
- 🔬 Principal AI Research Engineer at Technology Innovation Institute (TII), Abu Dhabi
- 🦅 Lead developer on the Falcon LLM team — Falcon 3, Falcon-H1, and Falcon-H1R model families (0.5B → 74B)
- 🚀 Large-scale, multi-node GPU training & inference with NeMo-RL, Megatron-LM, veRL, vLLM, and SGLang
- 🛠️ Previously AI Frameworks Architect at Intel — optimizing LLMs (LLaMA, BLOOM, GPT-class) on AI accelerators
- 📈 20+ years of overall industry experience
Pre-training · SFT · RL (GRPO · DAPO · DPO) · Reasoning & test-time scaling · Synthetic data generation · Long-context training · TP / PP / DP / FSDP · Context & sequence parallelism
| Year | Title | Link |
|---|---|---|
| 2026 | Falcon-H1R: Pushing the Reasoning Frontiers with a Hybrid Model for Efficient Test-Time Scaling | arXiv |
| 2026 | Falcon-H1-Tiny: A Series of Extremely Small, Yet Powerful Language Models | HF Blog |
| 2025 | Falcon-H1: A Family of Hybrid-Head Language Models Redefining Efficiency and Performance | arXiv |
| 2024 | Welcome to the Falcon 3 Family of Open Models | HF Blog |
| 2023 | Defect Classification for Integrated Circuits Contamination on Land Grid Arrays | IEEE |
| 2022 | Screening Deep Learning Inference Accelerators at the Production Lines | VLSI-2022 |
| 2021 | Identify and Localize COVID-19 Abnormalities on Chest Radiographs (MSc thesis) | ResearchGate |
- 💼 LinkedIn — linkedin.com/in/puneeshkhanna
- 🤗 Hugging Face — huggingface.co/puneeshkhanna
- ✉️ Email — puneesh.khanna83@gmail.com