Hi, I'm Chhalma

Architect and own end-to-end delivery of a production AI text detection system — from model development to cloud deployment.

• Developed RoBERTa-based classifiers using Hard Negative Mining, achieving 75% reduction in false positives.
• Designed and maintain cloud infrastructure on Microsoft Azure including CI/CD pipelines and containerised deployments via Docker.
• Implemented production components including authentication (Keycloak) and payment systems (Stripe).
• Designed prompts for synthetic data generation using LLMs for model fine-tuning.
• Planning agentic ML retraining pipeline using Claude API with tool use for automated model orchestration.

Engineered and optimised multi-stage benchmarking pipelines for DeBERTa V3-Large using PyTorch, ONNX, and TensorRT.

• Achieved up to 4× inference speedup through multi-batch optimisation across CPU and GPU environments.
• Analysed latency, throughput, and memory trade-offs across full-precision and quantised models.
• Containerised pipelines using Docker for reproducible deployments.

Built backend APIs and integrated AI features for an investor-founder matching platform.

• Developed NestJS APIs and SQLite database for platform functionality.
• Collaborated in Agile team using Slack for remote coordination.
• Integrated AI decision-making logic into matching algorithms.