ML Systems Engineering | Adaptive LLM Routing | Model Reliability
I work on adaptive AI systems that study confidence estimation, calibration, and cost–accuracy tradeoffs between small and large language models.
- Adaptive routing between small and large language models
- Confidence estimation and uncertainty-aware inference
- Calibration and overconfidence analysis in transformers
- Failure mode analysis in NLP systems
- Efficient inference under resource constraints
A task-adaptive routing framework that dynamically escalates queries between small and large models using uncertainty heuristics and systematic cost–accuracy benchmarking.
An empirical reliability study measuring calibration error (ECE), overconfidence rates, and temperature scaling effects across transformer models.
An interpretable classical ML pipeline for time-aware financial risk prediction with lead-time evaluation.
More details in pinned repositories.
Core ML — PyTorch · HuggingFace · scikit-learn
LLM Systems — Ollama · Transformers · Prompt Engineering
Evaluation — Calibration Metrics · Reliability Analysis
Data & Infra — Pandas · NumPy · SQLite · FAISS
Applications — Streamlit · FastAPI
Outside machine learning, I follow Formula 1, watch football, play competitive and narrative-driven games, unwind with music, and occasionally revisit anime.
LinkedIn: https://linkedin.com/in/theskybiz141
Email: aakashjsrindia@gmail.com
Precision matters — whether in model calibration or a last-lap overtake.