I'm Jayesh Suryavanshi.
I build LLM agents & GenAI
in production at .
About Me
I build LLM agents and the systems that make them work in production.
About Me
I'm a Machine Learning Engineer at eBay, where I build LLM agents and GenAI systems that run in production at marketplace scale. My work spans agentic services (LLM-generated summaries, multimodal analysis across text and images, MCP tool servers) and the classic ML behind them, like gradient-boosted models trained on large-scale data. Before eBay, I worked on ML systems for the financial industry at FINRA, and built data platforms at Datava and Infosys. I hold an MS in Computer Science from the University at Buffalo (SUNY). I care about shipping AI systems that are measurable, safe, and actually used in production.
Technical Skills
• LLM & Agentic Systems: Claude (Anthropic API), Azure OpenAI, Vertex AI, MCP (Model Context Protocol), agent orchestration, RAG, prompt engineering, LLM evals
• Machine Learning & NLP: PyTorch, TensorFlow, Hugging Face Transformers, XGBoost, spaCy, NLTK
• Programming Languages: Python, Java, SQL, Scala, Go, Bash
• Data & Distributed Systems: Spark, Hadoop, Hive, Kafka, Flink, ETL, Pandas, NumPy
• Cloud & CI/CD: AWS, Docker, Kubernetes, Terraform, Jenkins, Ansible
more ➜
Career 👨🏻💻
eBay Machine Learning Engineer, 3
I build production GenAI systems at eBay. My main focus is an LLM-powered service that turns complex, multimodal inputs (long text threads, images, OCR output) into structured summaries and signals, and exposes its capabilities to AI agents as tools over MCP. I also work on an internal multi-agent platform (agent registries, MCP server configuration, CI validation for agent packs) and build multi-stage XGBoost models on marketplace-scale data.
FINRA Software Engineer, Machine Learning
Built machine learning and AI systems addressing challenges in the financial industry.
Datava Software Engineer
Experienced Software Engineer proficient in Data Infrastructure contributing to the growth and success of fintech startup Datava.
SUNY Research Foundation Research Intern
Worked under the distinguished guidance of the renowned Prof. Dr. Bina Ramamurthy on Distributed Systems at SUNY Research Foundation, New York. Research Topic: A DeFi protocol that operates on a blockchain, enabling automated transactions between cryptocurrency tokens on the Ethereum network (Ropsten) without the need for traditional intermediaries.
Infosys Software Engineer, Data Platform
As a Software Engineer, I've developed scalable backends in Python and Java, created efficient data pipelines using Spark, AWS, and Jupyter Notebooks, and improved OTA update systems. I've built ETL processes to streamline data ingestion from varied data sources, enhancing data availability for data scientists and ML engineers. I've collaborated with ML engineers to build scalable workflows and automated ETL pipelines using CI/CD tools like Jenkins, Ansible, and Airflow, with Git for version control.
Internshala Software Engineer Intern
Ideated and created mockups, UML diagrams, and lean business plans for the Internshala Student Portal. Formulated the technical process flow for functionalities and working of online training system.
Education 👨🏻🎓
University at Buffalo, SUNY Master of Science (MS), Computer Science
University of Pune Bachelor of Engineering (BE), Computer Engineering
Acumen
LLM agents, ML models, and the systems that run them.
I build AI systems end to end: LLM agents and multi-provider GenAI services on top of solid ML models, data pipelines, and infrastructure. The goal is always the same: systems that hold up in production, where the decisions matter.
-
Building LLM Agents & GenAI Systems
I design and ship LLM-powered services end to end: agentic workflows that summarize and reason over real business data, multimodal analysis across text and images, PII redaction and content moderation, and MCP servers that let other AI agents call these capabilities as tools. I work across providers, from Claude (Anthropic), Azure OpenAI, and Vertex AI, calling multiple models in real time with structured, machine-readable outputs.
-
Applied ML & Predictive Modeling
I build predictive models that drive real product decisions, including multi-stage gradient-boosted cascades that refine their predictions as new signals arrive. Trained on distributed data at scale, and treated as products: feature pipelines, monitoring, and evaluation matter as much as the model itself.
-
Data Engineering at Scale
Models are only as good as the data behind them. I design and maintain pipelines on Spark, Hadoop, Hive, and Kafka that turn massive, messy datasets into reliable features and signals (batch and real-time) so both ML models and LLM systems have reliable inputs.
-
Cloud & Infrastructure
I run AI workloads on cloud infrastructure using AWS, Docker, Kubernetes, and Terraform, deploying and scaling services with an emphasis on reliability, observability, and cost. Infrastructure as code by default.
-
CI/CD & Automation for AI Systems
I automate the path to production with Jenkins, Airflow, and Git-based workflows, including CI validation and security scanning for agent configurations, and tooling that generates LLM-readable documentation for codebases. Shipping AI safely should be repeatable, not heroic.
Open Source
Code I have shipped into libraries other people depend on.
Four pull requests merged into maintained scientific Python libraries, each reviewed and accepted by the people who run them. One of them shipped as a named estimator in a release. Alongside two packages of my own on PyPI.
-
aeon · MADRID anomaly detector
A multi-length discord detector for time series, contributed as a new estimator to aeon and merged in September 2026. It shipped in the v1.6.0 release, whose notes read “New estimators, including the ExtendedIsolationForest and MADRID anomaly detectors”. I am listed in the project’s contributors file. PR #3702
-
PyOD · save and load round-trip coverage
-
category_encoders · GrayEncoder correctness fix
GrayEncoder.inverse_transformdid not recover the original values it had encoded, so a documented round trip silently returned the wrong answer. Fixed with a regression test and merged by the lead maintainer of scikit-learn-contrib/category_encoders. PR #499 -
scikit-dimension · uncentered lPCA
Added a
centeroption to the lPCA intrinsic-dimension estimator, so it can run on uncentered data where the origin carries meaning rather than forcing a recentering that changes the estimate. PR #41 -
Packages I maintain
graphspot is inductive graph anomaly detection with honest flat baselines and a core that does not require torch, which the comparable libraries do without declaring it. It ships flat baselines beside the graph models on purpose, because a detector on raw features is a genuinely hard baseline: on Reddit, the BOND benchmark has LOF outscoring every deep graph model. GeoSocialX maps the geography of geotagged social data and reports its coverage before it maps anything, because most posts carry no location at all. Both install from PyPI.
-
In review
Twelve further pull requests are open with maintainers, across scikit-learn, PyOD, river, imbalanced-learn, hiclass, pythresh, DeepOD and aeon. Four add new detectors, the other eight are correctness fixes. Open is not merged, so I count them separately.
Projects
Some of my notable projects I've worked on.
graphspot
Inductive graph anomaly detection (on PyPI as graphspot), with a core that installs and runs without torch. It ships flat baselines next to the graph models on purpose, because a detector on raw features is a genuinely hard baseline to beat. The thumbnail is real output: OddBall, unsupervised, scoring a neighbourhood of the Tolokers graph.
Lyft Motion Prediction for Autonomous Vehicles
Motion prediction on the Lyft Level 5 dataset, rebuilt as a study of what the task looks like under a single-laptop compute budget rather than a leaderboard run.
Triangulation: SQL Explorer
An IMDb-style movie/TV database you can query with free-form SQL, entirely in the browser: a real SQLite engine compiled to WebAssembly with a bundled sample dataset: no backend, no network. Rebuilt from a 2022 database-course project into a self-contained live demo with pre-built analytics views and 16 passing tests.
Beacon: P2P EV Charging
A peer-to-peer marketplace for sharing home EV chargers, built as an Ethereum dApp: a pay-as-you-go Solidity contract audited and hardened against fund-drain bugs (15 passing Hardhat tests), with a React + ethers v6 frontend where hosts list chargers and drivers pay per charge in ETH. UB blockchain-course project.
Social Unrest Forecasting
A leakage-controlled study of subnational unrest forecasting on ACLED, UCDP and GDELT data: rolling-origin backtests over 55 countries against persistence baselines the field rarely beats. Headline finding: news-text signals are informative but subsumed. They can substitute for curated event histories, but don't improve on them. A rigorous 2026 rebuild of my 2022 master's project.
GeoSocialX
A Python package (on PyPI as geosocialx) that maps the geography of X posts: fetch geotagged posts by radius via the X API v2, then surface hotspots and time trends.
Walleto: Full-Stack Digital Wallet
Led development of full-stack digital wallet app, enabling users to connect bank accounts, view transactions, pay bills, and earn cashback