Sabarish Bharadwaz Vadarevu
Verified Expert in Engineering
Computer Vision Systems Engineer and Developer
Hyderabad, Telangana, India
Toptal member since April 10, 2026
Sabarish is an ML systems engineer specializing in building reliable AI systems for real-world use. With a PhD in computational physics and experience leading large-scale CV platforms at Akridata, he focuses on data-centric AI, embedding pipelines, and agentic/LLM workflows. His work emphasizes system-level debugging, failure analysis, and the improvement of correctness and performance in complex AI pipelines.
Portfolio
Experience
- Linux - 15 years
- Research - 13 years
- Python - 13 years
- OpenCV - 12 years
- Computer Vision - 8 years
- PyTorch - 8 years
- Anomaly Detection - 3 years
- AI Agents - 2 years
Work Experience
Senior Manager, Data Science
Akridata Data Explorer
- Led the development of a data-centric computer vision platform spanning dataset management, labeling workflows, training, evaluation, and monitoring across 3,500+ ML work items.
- Developed agentic ML workflows integrating dataset operations, knowledge base creation, LLM-based analysis, training, and evaluation into cohesive pipelines.
- Implemented anomaly detection and multimodal reasoning systems combining CV models, embeddings, and LLM-based analysis.
- Designed embedding-driven dataset exploration tools enabling similarity search, clustering, and outlier detection on large-scale image datasets.
- Built ML-assisted annotation and autolabeling pipelines, accelerating dataset creation and enabling iterative model improvement.
- Introduced test-driven ML practices and CI/CD pipelines, improving experimentation speed and deployment reliability.
- Implemented 650+ improvements across ML infrastructure and CV systems, improving model development efficiency and system reliability.
Research Fellow
The University of Melbourne
- Conducted high-fidelity numerical simulations to study turbulence dynamics in high Reynolds number flows, focusing on impulse response behavior.
- Identified coherent vortex–streak structures in simulated flow fields, providing evidence for fundamental building blocks of turbulence.
- Developed computational models to analyze linear and nonlinear mechanisms governing flow instability and structure formation.
- Designed and executed large-scale numerical experiments, translating theoretical concepts into validated computational results.
Project History
Data-centric Computer Vision Platform | Akridata DataExplorer
https://akridata.ai/data-explorer/I led development across dataset management, embedding-driven exploration, labeling workflows, training, evaluation, and monitoring. I designed systems for similarity search, clustering, and outlier detection on large-scale image datasets, enabling teams to systematically identify failure modes. I also implemented ML-assisted annotation and autolabeling pipelines to accelerate dataset creation and iteration cycles.
The platform unified data, models, and workflows into a cohesive system that continuously improves model performance in real-world deployments.
Agentic Anomaly Detection and Knowledge System
I designed workflows that connect dataset exploration, model inference, and knowledge generation, allowing the system to learn as users define and refine their problem. I also built orchestration pipelines that transform fragmented ML steps into a cohesive, adaptive system, enabling faster insight generation and decision-making in complex, unstructured data environments.
I fully ideated and executed this as an advanced-tech prototype.
Defect Synthesis System
https://docs.akridata.ai/docs/datagenVision-language Anomaly Detection Pipeline
https://akridata.ai/vision-assist/Systems that combine visual embeddings with language-based reasoning were built to detect, classify, and explain anomalies beyond traditional supervised approaches. I enabled flexible querying and semantic search over image datasets, improving detection of edge cases and previously unseen failure modes. I also integrated these pipelines into production workflows, allowing teams to analyze anomalies, generate insights, and iteratively improve model performance using both visual and textual signals.
As the team lead, I also contributed individually to enhance team output.
Boardroom Simulator
Initial implementations using CrewAI exposed limitations in common multi-agent patterns, including shallow task delegation and weak discussion control. A later LangGraph-based architecture introduced explicit routing, iterative critique/revision loops, persona-specific subagents with long-term memory, and state compaction to maintain coherence across extended deliberation cycles. The system used a lightweight MCP server for shared contextual resources rather than a full RAG, since the knowledge base consisted of a small set of research reports and strategic documents. Experimental web search integrations (SearXNG, Tavily, Exa) were evaluated for live context augmentation, but ultimately excluded due to noise and reduced discussion quality.
Education
PhD in Aerospace Engineering
University of Southampton - Southampton, UK
Master's Degree in Aerospace Engineering
Indian Institute of Technology - Madras (IITM) - Chennai, India
Bachelor's Degree in Aerospace Engineering
Indian Institute of Technology - Madras (IITM) - Chennai, India
Certifications
IBM RAG and Agentic AI Professional Certificate
IBM | via Coursera
Skills
Libraries/APIs
PyTorch, OpenCV, Pydantic, TensorFlow, Hugging Face Transformers, Asyncio, OpenAI API, React, OpenAI Assistants API, Node.js, llama.cpp
Tools
Git, Visual Language Models (VLMs), AI SDK, MATLAB, Claude, Codex, Claude Agent SDK, Azure ML Studio, Azure Machine Learning, Blender, ParaView, OpenFOAM, n8n, NVIDIA Jetson, NX CAD, SOLIDWORKS
Languages
Python, Markdown, SQL, TypeScript, C++
Paradigms
Anomaly Detection, Synthetic Data Generation, Automation, Testing, Unit Testing, Rule-based Programming, Model Context Protocol (MCP), DevOps
Frameworks
LlamaIndex, Agentic Frameworks, Flask, LangGraph, AutoGen
Platforms
Linux, Amazon EC2, Amazon Web Services (AWS), Ollama, Google Cloud Platform (GCP), Vertex AI, Harness, LangSmith, Docker, Azure, Databricks, CrewAI
Storage
PostgreSQL, Data Pipelines, Elasticsearch, Datadog
Other
Embedding Models, Vector Search, Research, Numerical Methods, Simulations, Fluid Dynamics, Scientific Computing, Experimental Design, Linear Algebra, Applied Mathematics, Computer Vision, Clustering, Data Labeling, AI Agents, Vision-Language Models, ML Pipelines, Artificial Intelligence (AI), AI Pipeline, Large Language Models (LLMs), Image Processing, Retrieval-augmented Generation (RAG), Prompt Engineering, Scalable Vector Databases, Machine Learning, Deep Learning, FAISS, AI Integration, API Integration, Optical Character Recognition (OCR), Multimodal Models, Hugging Face, OpenAI, Coding, Architecture, Leadership, SDKs, Team Leadership, Agentic RAG Systems, Clustering Algorithms, AI Tools, Machine Learning Operations (MLOps), Software Development Lifecycle (SDLC), System Architecture, Benchmarking, Hyperparameter Tuning, OpenAI GPT-4 API, Computer Vision Algorithms, Convolutional Neural Networks (CNNs), Residual Neural Networks (ResNets), Data Science, Gemini, Large Language Model Operations (LLMOps), Machine Learning (ML) APIs, Fine-tuning, Workflow Automation, AI Agent Orchestration, LLM Integration, Semantic Code, Cloud, Semantic Search, AI Design, Algorithms, Generative Artificial Intelligence (GenAI), Agentic Workflow Design, FastAPI, MLflow, Multistage LLM Chains, Diffusion-based AI Models, Diffusion Models, GPU Computing, Agentic AI, Agentic AI Systems, DreamBooth, Video Processing, RAG Architecture, RAG Pipelines, LangChain, Vector Databases, Pinecone, Recommendation Systems, Decision Modeling, Document Processing, AI Copilots, Natural Language Processing (NLP), RAG Systems, Predictive Modeling, CI/CD Pipelines, Facial Recognition, Facial Tracking, Video Surveillance, Edge Computing, Applied Physics, DNS, Visualization, Engineering, Modeling, ANSYS, Systems Thinking, Data Analysis, Dynamic Systems, Workflows, Dash, Data Augmentation, Decision Trees, Multi-agent Systems, Hooks, Gemini API, Anthropic, LiteLLM, Azure Databricks
How to Work with Toptal
Toptal matches you directly with global industry experts from our network in hours—not weeks or months.
Share your needs
Choose your talent
Start your risk-free talent trial
Top talent is in high demand.
Start hiring