Abdul Basit Anees

A call center studio for building voice agents, and a voice customer-service agent for a client that keeps getting better from its own evals. I also developed Koder, aiXplain's coding agent.

I mine test cases from production logs, write more from plain-text guidance, and run each one several times. Then a prompt optimization loop built on GEPA improves the prompt within a dollar and iteration budget, and a held-out set checks the result.

Text-driven image editing with StyleGAN and CLIP, published at SIGGRAPH Asia and in ACM Transactions on Graphics. More recently, LLM-driven candidate assessment at EACL 2026, the ranking method behind aiXplain's HR manager.

From 2018 I wrote models from scratch: linear and logistic regression, KNN, SVM, RNNs, transformers and GANs. In 2020 I moved to PyTorch and TensorFlow and kept building from scratch on the side. Since 2023 I have worked mostly on LLMs, agents and retrieval.

Python and PyTorch. Qdrant and Neo4j. FastAPI. Docker and Kubernetes on AWS, with Datadog for tracing.

Every retrieval change gets benchmarked against what's already in production, on recall, latency and cost, before it goes live.

aiXplain since 2023. Before that: Radity, the KUIS AI Center at Koç University, and internships at Fraunhofer IIS, Visover and Argedor.