Abhishek Vats
Verified Expert in Engineering
Machine Learning Developer
Delhi, India
Toptal member since June 15, 2026
Abhishek is an AI and ML expert with 6+ years of experience, recently focusing on LLMs, generative AI, agent development, and computer vision. He has built scalable LLM-based systems and led multicultural teams across different countries. His work spans insurtech, edge AI, geospatial intelligence, sports tech, and climate tech, blending technical depth with strategic leadership.
Portfolio
Experience
- Python - 6 years
- Computer Vision - 6 years
- Deep Learning - 5 years
- PyTorch - 5 years
- FastAPI - 5 years
- AWS IoT - 4 years
- AI Agents - 2 years
- LangGraph - 2 years
Preferred Environment
Linux, Python, AWS IoT, LangGraph, Langfuse, PyTorch
The most amazing...
...system I've built is a full-stack AI agentic app where creators live on as interactive audio and video personas their audience can have real conversations with.
Work Experience
Lead AI Engineer
Stupa Sports
- Built a white-label customer-support RAG assistant over PRDs, code repos, and API docs, deployed per-tenant to replace recurring customer training meetings. Used chunking and indexing into Qdrant (BGE embeddings) and Cohere reranking.
- Built a conversational analytics layer for tournament organizers to query their data in natural language and get charts/plots on demand.
- Built an automated line-calling system for professional badminton, replacing manual line judges with a multi-camera vision pipeline.
- Solved high-velocity small-object tracking (shuttle at 400+ km/h) using RF-DETR with Kalman-filter motion smoothing and epipolar constraints for cross-view occlusion handling.
- Productionized Stupa Cast, a real-time analytics system for live table tennis broadcasts, using multiple specialized networks ( Pose, Action Recognition, Detection) , ingesting live video to emit on-air stats and rally annotations.
Senior ML Consultant
IBM
- Reimagined boring old JSON weather APIs into smart LLM-enhanced APIs that transform raw forecasts into personalized recommendations (cycling conditions, ski trip planning) and TV broadcast-ready audio forecast using AI based TTS.
- Designed production-grade agentic architecture using LangGraph orchestration, self-hosted LiteLLM gateway (multi-provider passthrough), custom tool chains, and Pydantic-enforced schemas for reliable structured outputs.
- Deployed the full stack on AWS ECS with Dockerized microservices and auto-scaling inference containers behind an ALB.
Senior ML Engineer
Keywords Studios
- Built an in-house secure and private alternative AI assistant to Claude and ChatGPT for Keywords Employees. Used Llama3.1 as the LLM of choice and an intuitive front end. Added a company-wide knowledge base from codebases to employee Slack threads.
- Worked on exploring the latest research in text-to-3D asset generation for game development.
- Designed a multi-agent LLM system to create 2D floorplan layouts from mere text prompts.
Senior ML Engineer
Shibumi.AI
- Built an end-to-end pet toonification pipeline, enabling users to upload photos of their pets and generate Disney-style cartoon renditions while preserving identity-defining features (breed, coloring, distinguishing marks).
- Fine-tuned stable diffusion via DreamBooth on user-submitted pet images to teach the model a subject-specific token for each pet, then combined this with style prompts and conditioning to transfer a consistent Disney/toon aesthetic.
- Deployed the fine-tuned models on AWS SageMaker inference endpoints to serve the user-facing app, handling the full workflow from image capture and preprocessing through model inference to output delivery.
ML Engineer
SiteRecon Inc
- Designed an automated property mapping system using satellite images for landscaping businesses in the US, speeding up their quote process and removing the need for physical labor and site visits.
- Trained custom U-Net architectures for fine-grained semantic segmentation of landscape features, augmented with domain-specific post-processing heuristics (class adjacency rules, area thresholds) to suppress false positives and refine masks.
- Built an automated data labeling and active learning pipeline using Label Studio, integrated with model-in-the-loop annotation suggestions, reducing manual labeling effort by 60% and continuously improving segmentation accuracy.
Experience
CreatorTwin
Badminton Line Calling
https://youtu.be/HRQMylEPYOYWeather Agent
Whether it's telling a cyclist the exact window to ride before wind picks up, or generating a broadcast-ready audio forecast a TV station can air directly, the system thinks rather than just reports.
Education
Bachelor's Degree in Computer Science and Engineering
University School of Information, Communication, and Technology - Delhi, India
Skills
Libraries/APIs
PyTorch, TensorFlow, Pydantic
Tools
You Only Look Once (YOLO), Git, Sahi, Okapi BM25
Languages
Python, TypeScript
Platforms
Linux, AWS IoT, Langfuse, LiveKit, Ollama
Frameworks
LangGraph, LlamaIndex, Mem0
Paradigms
Model Context Protocol (MCP)
Storage
MongoDB, PostgreSQL
Other
U-Net, Computer Science, Machine Learning, Computer Vision, Artificial Intelligence (AI), Pose Estimation, Retrieval-augmented Generation (RAG), LangChain, RAG Pipelines, Large Language Models (LLMs), Agentic AI, AI Architecture, Prompt Engineering, NVIDIA TensorRT, Motion Capture, Agentic RAG Systems, Vector Databases, Model Evaluation, Data Science, Speech-to-Text (STT), Text-to-Speech (TTS), AI Hallucinations Management, CodeLlama, LoRa, QLoRA, Qdrant, Cohere, RAGAS, Deep Learning, FastAPI, LiteLLM, Geospatial Analytics, LLM Integration, AI Agents, Convolutional Neural Networks (CNNs), PipeCat, ElevenLabs Solutions, HeyGen, WebSockets, Tracking, Object Detection, Semantic Segmentation, 3D Reconstruction, Amazon Bedrock AgentCore, Stable Diffusion, DreamBooth, Open-source LLMs
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