JAYAKRISHNAASubramaniyanRavichandran
Engineering intelligence
I design production AI that can reason, remember, recover and earn trust—from clinical copilots to self-healing infrastructure.
Selected systems · 2023—2026
Each system is designed around the harder questions: evidence, memory, failure, cost and human control.
01 · Completed
02 · Research work
03 · In process
04 · Architecture in 3D
Every project as a frontend-to-backend flow you can rotate and tilt. Gold travels in with the request, white comes back with the answer.
Point of view
Models are only one layer. Useful AI also needs orchestration, evidence, memory, evaluation, observability and a safe way to fail.
Ground before generating. Retrieval and protocol evidence come before confident output.
State is a product decision. Memory is designed for continuity, cost and safe rollback.
Autonomy needs boundaries. Models can propose; deterministic policy controls the blast radius.
Working vocabulary
LangGraph · RAG · LLM evaluation · Prompt systems · Conversational AI
NLP · Computer vision · Classification · Forecasting · Recommendations
Python · FastAPI · SQL · PySpark · Redis · MongoDB · OpenSearch
AWS Bedrock · SageMaker · Lambda · Docker · Kubernetes · CI/CD · MLflow
Experience
Chennai, India
2025 — 2026
Cavin Infotech
Designed and shipped three production LangGraph systems across veterinary triage, clinical decision support and enterprise knowledge retrieval.
2023 — 2025
KMU Techgrid Solutions
Owned machine-learning lifecycles across finance, computer vision and conversational AI, with production deployments on AWS.
2022 — 2023
KMU Techgrid Solutions
Turned operational data into decision-ready reporting and supported early ML and GenAI prototypes for business teams.
Earlier explorations
Deep learning · Image classification
Predictive modelling
Computer vision · Biometrics
Feature extraction · Authentication
Edge AI · Raspberry Pi
B.E. Electronics & Communication Engineering
PRIST University · 2013—2017
Available for the right problem