Background / current direction
I am an AI researcher and systems architect based in Kolkata, India. As the founder of KindSynapse AI and a core maintainer at Elixpo, I design compute-efficient neural models, distributed caching layers, and zero-GPU edge inference engines. I also conduct AI research with Pollinations.ai and undergraduate research at IEM Kolkata.
The practical direction
My work focuses on removing computational bottlenecks in modern deep learning. At age 17, I built Azinet, a zero-dependency neural network library from scratch in pure Java, and qualified for the interview stage of PAIR FABRIC where I was interviewed by a joint Harvard-MIT researcher. Today, I direct KindSynapse AI, co-authored the 3-tier caching architecture behind OreoLook (arXiv:2609.05463), and won the Gold Medal at IEMPHYS-26 for ARGUS, a linear-complexity vision transformer for CERN particle tracking.
I am open to conversations about AI systems research, extreme quantization kernels, linear-time attention mechanisms, and open-source infrastructure collaborations.
Start a conversation ↗Experience and responsibility
Current and past work is recorded through the responsibility held and the system or community served.
June 2025 - Present
Directing an open-source AI collective advancing resource-efficient neural architectures, 1-bit quantization kernels, and high-throughput zero-GPU edge inference engines. Ecosystem scaled to 840,000+ total visits, 3,200+ combined community likes, paying international clients across the US, Italy, and Singapore, and formal recommendation by the Government of Germany (KI:Connect NRW).
March 2026 - Present
Core maintainer for search.elixpo / OreoLook, building distributed 3-tier caching hierarchies (hot-cache, semantic vector deduplication with Redis and Qdrant, cold archival) and co-authoring the formal research publication arXiv:2609.05463. Awarded Elixpo Recognition of Contribution (ELX-APP-2026-006).
March 2026 - Present
Contributing to the core AI research group; published PollinationsAI's first-ever formal academic research paper on low-latency CPU caching architectures (arXiv:2609.05463) and co-founded the PollinationsAI Research (P-Research) organization on Hugging Face.
March 2026 - Present
Leading machine learning architecture for an early-stage venture creating automated psychometric assessment platforms; engineering low-latency neural pipelines and privacy-preserving psychological analysis extractors.
2025 - Present
Designed the official chapter logo and digital/media branding assets. Serving as Point of Contact for technical systems, research initiatives, media production, and industry relations.
