Tuhin Sharma
Co-Founder & AI Architect
Verified Proof of Work Artifacts
2 items catalogedEach artifact below represents an authenticated research publication, production code repository, or technical architectural framework directly authored or co-created by Tuhin Sharma. Every entry undergoes editorial source verification.
Small Language Model (SLM) Fine-Tuning and Inference Optimization
Engineered domain-specific distillation pipelines compressing 70B parameter models into 3B-7B parameter edge models for on-premise enterprise deployment with zero data leakage.
SLMs have lower zero-shot generalization on tasks outside the fine-tuning training distribution.
Enterprise Agent Routing with Model Cascades and Latency Guardrails
Architected semantic routing proxies that direct routine classification to cheap SLMs while escalating complex reasoning tasks to frontier LLMs, slashing cloud compute costs by 65%.
Threshold calibration requires ongoing monitoring to prevent misrouting edge-case queries to smaller models.