# Naman Chawla > AI/ML Engineer at Coding Blocks in New Delhi, India. Builds AI products end-to-end on top of large language and generative models: agents, assistants, voice and video AI, and the deployment and GPU infrastructure that runs them in production. Outside of work: research papers and side projects on reinforcement learning, quantization, and distillation. ## Pages - [Home](https://palindromerice.github.io/): About Naman, with contact and social links - [Writing](https://palindromerice.github.io/writing/): Articles, published research papers and LinkedIn posts - [Projects](https://palindromerice.github.io/projects/): Personal ML projects with code ## Articles - [The Distant Crush Neuron: Single-Neuron Steering of Social Warmth in Qwen3.5-4B](https://palindromerice.github.io/writing/qwen-neuron/): Steering one MLP neuron at inference makes a chat character cooler or warmer, checked by blind raters against random-neuron controls, 2026 - [I Keep Losing My TV Remote, So I Turned My Mac Into One](https://palindromerice.github.io/writing/lgtv-remote/): A Mac menu-bar remote for LG webOS TVs over the local network, 2026 - [Agent-Assisted Reconstruction of Quake II Gameplay Code](https://palindromerice.github.io/writing/quake2-reconstruction/): Coding agents rebuild Quake II's gameplay module from stripped binaries; 99.0% of testable functions match and the module is playable, 2026 - [Small Models as Agents: SFT and GRPO on a Browser Task](https://palindromerice.github.io/writing/browser-agent-sft-grpo/): Qwen3.5-0.8B goes from 16% to 84% success on a live browser task, 2026 - [CAPL: Context-Aware Preference Rules for Steering a Frozen LLM](https://palindromerice.github.io/writing/capl/): Context-scoped preference rules steer a frozen generator, 2026 ## Papers - [ClickGuard: A Trustworthy Adaptive Fusion Framework for Clickbait Detection](https://arxiv.org/abs/2604.07272): arXiv, 2026 - [Skin Cancer Classification Using Ensemble of Deep Learning Models and Transfer Learning](https://doi.org/10.1007/978-3-032-30438-4_45): LNNS, 2026 - [Decoding Sarcasm: Leveraging GloVe and LSTM for Accurate Sentiment Understanding](https://doi.org/10.1201/9781003654483-68): CRC Press, 2025 - [Hybrid Model Approaches Toward Movie Recommendation Systems and Their Comparisons](https://link.springer.com/chapter/10.1007/978-981-99-6547-2_49): LNNS, 2024 - [Unveiling Emotions Through Sentiment Analysis](https://papers.ssrn.com/sol3/papers.cfm?abstract_id=5023927): SSRN - [Gesture-Based Touchless Operations: MediaPipe, OpenCV](https://dergi.neu.edu.tr/index.php/aiit/article/view/799): AIIT ## Projects - [Qwen hot-cold neuron](https://github.com/palindromeRice/qwen-hot-cold-neuron): One MLP neuron that turns a chat character warmer or cooler at inference, with no weight changes - [LG TV Remote for Mac](https://github.com/palindromeRice/lgtv-remote-reverse-engineer): Menu-bar remote for LG webOS TVs with a trackpad pointer and power-on from standby - [CAPL](https://github.com/palindromeRice/capl): Context-aware preference rules for steering a frozen LLM - [Quake II reconstruction](https://github.com/palindromeRice/quake2-llm-decompile): Agent-assisted reconstruction of Quake II gameplay code with a differential test harness - [openEnvRL](https://github.com/palindromeRice/openEnvRL): SFT and GRPO for a small model acting in a live browser - [Single-Bit Quantization of Transformers/BERT](https://github.com/palindromeRice/SingleBitLLMs): Benchmarks single-bit/ternary quantization strategies inspired by BitNet b1.58 on BERT - [T5 Knowledge Distillation for Summarization](https://github.com/palindromeRice/Knowledge_Distillation_T5): Distills a fine-tuned T5-Base summarizer into T5-Small - [Mixture-of-Experts Prompt Router](https://github.com/palindromeRice/Moe_demo): Routes prompts to specialized expert models via sentence-embedding similarity ## Contact - [Email](mailto:namanchawla2404@gmail.com) - [GitHub](https://github.com/palindromeRice) - [LinkedIn](https://www.linkedin.com/in/naman-chawla-0563032b5/) - [X](https://x.com/palindromeRice) - [Google Scholar](https://scholar.google.com/citations?user=Hq_k3_wAAAAJ&hl=en)