Avishek Biswas

SaaS for AI-assisted paper study. Agentic search across 70,000+ Arxiv papers. Personalized recommendations, summaries, and chat.

• Solo-built full stack: NextJS, background jobs, Postgres, custom ML deployment, context engineering, agentic systems
• 1,200 users in 4 months, 1,900+ papers studied, 6,500+ queries answered

AI consumer behavior simulation engine. Simulates millions of AI agents at enterprise scale. Backed by KPMG, SXSW finalist. Patent-protected.

• Led multi-agent LLM systems simulating social media interactions in an RL-based sandbox
• AI personas study news patterns, post opinions, and influence each other - pivotal in Series-A funding
• Designed evaluation frameworks for generative text quality across multiple metrics

Agentic AI platform for enterprise customer experience. Powers brands like Jos. A. Bank, Lane Bryant.

• Primary author of client-side lightweight LMs (pre-ChatGPT) for call-center autocomplete - 40% faster response times
• Built Transformers-assisted embedding space to model live chat sessions and generate alternative responses
• Founding data scientist - owned full ML workflow: data prep, MLOps, research, training, deployment, user studies

• Deep RL agents for physics-based character control - Python, TensorFlow, Stable Baselines. Published at ACM MIG
• TA for Computer Graphics, AI, and Deep RL. Simultaneously RA + TA in 3 of 4 semesters

• Application security, patching, solution upgrades, and migration support (Oracle Apps DBA)
• First job - developed leadership, teamwork, and accountability