pragathi
@Pragathi
Pragathi Jayaram
Applied AI systems consultant · Product engineer · Founder of Anovi
About
I build software, intelligent systems and products, and have done that for about 10 years: embedded automotive diagnostics at Bosch, computer vision and multimodal AI at Mercedes-B... Show more
Language & Tools
Languages & core: Python · TypeScript · embedded C/C++ · SQL
Product & backend: FastAPI · PostgreSQL · React Native · React · Node.js · Docker
Applied AI & LLM: Vertex AI · Claude and Gemini APIs · structured output and prompt design · retrieval and citation grounding · evaluation and benchmarking
ML & signals: computer vision · transformers · multimodal and sensor fusion · explainable AI · TinyML and on-device quantisation · NumPy · pandas
Product & backend: FastAPI · PostgreSQL · React Native · React · Node.js · Docker
Applied AI & LLM: Vertex AI · Claude and Gemini APIs · structured output and prompt design · retrieval and citation grounding · evaluation and benchmarking
ML & signals: computer vision · transformers · multimodal and sensor fusion · explainable AI · TinyML and on-device quantisation · NumPy · pandas
Currently Exploring
Retrieval and citation design for nutrition evidence, built to refuse questions the sources cannot answer. Preparing Anovi's Android release. Cycle-aware planning, scoped to stay non-medical.
Achievements
Built and shipped Anovi solo, schema to store listing, approved on its first App Store submission.
LLMs in production where being wrong costs something: free-text meal logs to tructured ingredients, multi-objective recipe ranking.
Real-time multimodal AI at Mercedes-Benz Tech Motion: gesture, gaze, pose and transformer-based speech systems.
Design evaluation environments and grading criteria for coding-focused reinforcement learning benchmarks.
MSc thesis, TU Munich: radar hand-gesture pipeline at 97% accuracy on an ARM Cortex-M0.
LLMs in production where being wrong costs something: free-text meal logs to tructured ingredients, multi-objective recipe ranking.
Real-time multimodal AI at Mercedes-Benz Tech Motion: gesture, gaze, pose and transformer-based speech systems.
Design evaluation environments and grading criteria for coding-focused reinforcement learning benchmarks.
MSc thesis, TU Munich: radar hand-gesture pipeline at 97% accuracy on an ARM Cortex-M0.
Random Dev Quote
The interesting problem is rarely the model.
Contributions
Articles
chevron_left
