Jayashree Devanarayanan
AI Engineer
4+ years exp
4+ years ML exp
Python
SQL
Data Pipeline & Feature Engineering: 4/10
Active 1 day ago
Invite to interview
Download CVCV
Overview
Technical skills
Timeline
Roles
Overview
Computer vision-focused ML engineer (early-career) building classical feature-based image classifiers and simple desktop/web applications. The strongest proven skill is classical CV feature extraction and end-to-end integration into a Flask service, evidenced by main.py functions bg_sub and feature_extract together with app.py upload route. There is little evidence of reproducible experiment pipelines, production deployment hardening, extensive testing, or efficiency/scale engineering in public code.
Technical skills
Languages
2
Python
SQL
AI/ML
17
LangChain
LlamaIndex
LoRA
XGBoost
PEFT
Airflow
Hugging Face
AI Agents
RAG
Model Context Protocol
Vertex AI
Pandas
OpenCV
Scikit-learn
LLM
CrewAI
Transformers
DevOps
7
GitHub
Git
GitHub Actions
Docker
Terraform
GCP
Kubernetes
Databases
4
Google BigQuery
Milvus
Oracle
Redis
Other
17
ETL/ELT
SQLAlchemy
FastAPI
Tekton
BigQuery
FAISS
Apache Kafka
Google Cloud Run
Looker
Splunk
Rest API
Tableau
Postman
CI/CD
Computer Vision
Prompt Engineering
Debian
Timeline
AI Engineer
•
Middle
Latentview Analytics
•
Full-Time
Built agentic AI workflows on Apache Airflow to automate root-cause analysis for DAG failures, improving recovery time. Standardized CI/CD migration from Tekton to GitHub Actions using AI agents and implemented automated data contract enforcement with Dataplex to improve pipeline reliability. Developed and deployed AI/ML pipelines on Vertex AI with MLOps practices and supported container security and infrastructure maintenance using Docker and Terraform.
AI Agents
Airflow
Tekton
GitHub Actions
Vertex AI
Docker
Terraform
GenAI Engineer
•
Middle
Walmart Global Tech India
•
Full-Time
Developed a POS system assistant using Model Context Protocol (MCP), agentic AI, and RAG to deliver real-time troubleshooting. Built retrieval systems with LangChain, LlamaIndex, and Milvus to improve search accuracy and grounding. Created predictive maintenance using XGBoost and PCA, and implemented scalable GCP ETL/ELT pipelines with BigQuery and Cloud Composer; fine-tuned LLMs with Hugging Face using LoRA/PEFT to reduce inference latency.
Model Context Protocol
AI Agents
RAG
LangChain
LlamaIndex
Milvus
XGBoost
PEFT
LoRA
Hugging Face
Google BigQuery
Airflow
ETL/ELT
Middle AI/ML Engineer
Confidence: Medium ML Engineer
Computer vision-focused ML engineer (early-career) building classical feature-based image classifiers and simple desktop/web applications. The strongest proven skill is classical CV feature extraction and end-to-end integration into a Flask service, evidenced by main.py functions bg_sub and feature_extract together with app.py upload route. There is little evidence of reproducible experiment pipelines, production deployment hardening, extensive testing, or efficiency/scale engineering in public code.
Model Architecture & Training
3/10
How well models are designed and trained
Classical model training using scikit-learn SVM and GridSearchCV is present but there are no custom architectures, advanced training loops, or experiment tracking.
Data Pipeline & Feature Engineering
4/10
How data is prepared for models
Concrete handcrafted feature engineering and label construction from filenames are implemented; preprocessing and texture/shape/color features are explicitly computed.
Evidence
Experimentation & Evaluation
2/10
How results are measured and tested
Basic experimentation exists via GridSearchCV and cv_results_, but there is no systematic evaluation pipeline, no reproducible experiment tracking, and no test/validation reporting.
Evidence
MLOps & Deployment
3/10
How models are shipped to production
A simple Flask service integrates the CV pipeline and SQLAlchemy model for lookup, which demonstrates basic serving but lacks model serialization, versioning, or production-grade deployment practices.
Computational Efficiency
1/10
How efficiently computing resources are used
There is almost no evidence of computational efficiency work; some parts are computationally naive (pixel loops) and there is no GPU/quantization/batching optimization or profiling.
Research Depth & Innovation
2/10
Depth of research and new ideas
Applies established techniques (Haralick features, contour moments) showing domain knowledge in classical CV but no novel research, ablations, or reproduced paper implementations.
Expertise
Computer Vision & Image Analysis• Middle
Technologies
Python• since 2024 • Middle
SQL
Redis
LangChain• since 2022
Splunk
Terraform• since 2025
Airflow• since 2022
GCP
Oracle
Milvus• since 2022
OpenCV
FAISS
LlamaIndex• since 2022
LoRA• since 2022
Model Context Protocol• since 2024
SQLAlchemy
XGBoost• since 2022
GitHub Actions• since 2025
Vertex AI• since 2025
Debian
Scikit-learn
Prompt Engineering
Computer Vision
AI Agents• since 2022
PEFT• since 2022
CI/CD
Transformers
Pandas
Git• since 2005
Docker• since 2025
Kubernetes
CrewAI
LLM
RAG• since 2022
Tekton• since 2025
Google BigQuery• since 2022
Google Cloud Run
Hugging Face• since 2022
GitHub• since 2008
BigQuery
Machine Learning• mentioned only
Recommendations
- Develop small-to-medium classical computer vision prototypes that use handcrafted features and lightweight models and include model serialization (pickle/ONNX) and unit tests.
- Harden web serving by adding model persistence, input validation, safe file handling, and a clear deployment path (containerization and simple CI).
- Invest in basic experiment tracking and evaluation (train/val splits, reproducible runs, metrics logging with CSV or lightweight W&B/MLflow) and add simple performance profiling to remove pixel-wise Python loops.
Repositories
The developer's experience in this domain has been verified based on AI analysis of the following repositories: