Chintan
Gadhiya

Modular multi-agent AI system using CrewAI for short- and long-term stock prediction. Unified API routing for five LLM providers. Achieved 30% increase in trading accuracy with probability-based signals.

Inference pipeline using Trusted Execution Environment isolation to protect prompts and responses. Encrypted request handling, privacy-preserving logs, and mixed-precision benchmarks.

End-to-end audio intelligence pipeline, speech-to-text, sentiment scoring, call summarisation, and outcome extraction, with REST APIs for batch and real-time processing.

Low-latency WebSocket chat streaming LLM responses token-by-token. Asyncio concurrency supports simultaneous sessions with configurable prompts and session-state persistence.

Synthetic data pipeline injecting typos, malformed JSON, semantic noise, and adversarial phrasing. Trained compact LLMs via QLoRA with full robustness evaluation metrics.

Benchmarking suite for CNN and Transformer workloads. Applied torch.compile, mixed precision, and gradient checkpointing to generate performance optimisation reports.

Real-time multi-face recognition with anti-spoofing and confidence thresholding. Full-stack dashboard for attendance logs, report export, and user management.

Multi-module web platform with React frontend and Flask backend. Web Speech API and GPT-based NLP for interactive pronunciation practice and personalised feedback.

Lightweight Flask-based system capturing prompts and LLM responses in real time. Analysis scripts surface word frequency, response length, and failure modes for post-training workflows.