Available for new opportunities · USA
I design and ship Generative AI & machine learning systems: RAG pipelines, LLM-powered assistants, and predictive models that turn messy enterprise data into fast, trustworthy answers. 4+ years shipping production AI that measurably cuts manual work.
01 · About
I'm an AI/ML Engineer with 4+ years of experience building machine learning, NLP, and Generative AI solutions that improve document retrieval, issue classification, knowledge discovery, and decision support for enterprise teams.
My day-to-day spans the full lifecycle, from data cleaning and feature engineering, through model training and evaluation, to shipping FastAPI inference services and RAG pipelines that real teams depend on. I work closely with product, QA, and data teams to translate ambiguous operational problems into validated, production-ready AI features, and I care as much about measurable business impact as about model accuracy.
Recently I've been deepening my cloud & MLOps practice: containerizing inference services, deploying LLM workflows on Azure OpenAI, and building the monitoring and evaluation habits (MLflow tracking, drift checks, A/B validation) that keep models reliable after launch, not just at demo time. I also track where the field is moving next, things like LLM fine-tuning (LoRA/PEFT) and agentic, tool-calling workflows with LangGraph and multi-agent orchestration, so the stack I build with stays current, not just production-stable.
LangChain / LlamaIndex pipelines, embeddings, vector search, and prompt engineering that turn documents into answers.
Classification, clustering, and recommendation systems tuned with rigorous evaluation: precision, recall, F1, ROC-AUC.
Azure OpenAI, containerized FastAPI services, model tracking with MLflow, and CI/CD-driven deployment workflows.
Own AI features end-to-end: defining acceptance criteria with product/QA, validating outputs, and driving adoption.
02 · Skills
Core ML/NLP/GenAI tooling, plus the cloud & MLOps practices I use to get models into production and keep them healthy.
03 · System Design
A representative pipeline I design and productionize, from raw documents to a validated, cited answer served over a REST API.
Click any stage for what it actually does & where I've shipped it
FastAPI inference endpoints behind REST APIs, built for both real-time and batch prediction so internal apps can consume model output directly.
Metadata filtering + semantic search to cut irrelevant results, with generated answers validated against source documents before rollout.
Cross-validation, precision/recall/F1/ROC-AUC, and A/B testing feedback loops used to validate model changes before release.
04 · Infrastructure
Beyond the RAG pipeline above: how the client, serving layer, AI/ML orchestration, data layer, and cloud/MLOps infrastructure fit together end-to-end.
I keep the detailed infrastructure diagram gated to real conversations. If you're a recruiter or hiring manager, send a quick message and I'll personally unlock it for you.
That code doesn't match. Try again, or request access above.
Access stays open for 5 minutes, then locks again automatically.
The full architecture diagram and global footprint map are ready to view for the next 5 minutes.
Streamlit & Gradio prototypes, REST clients consuming model output directly.
FastAPI inference endpoints for real-time and batch prediction, plus REST APIs for internal app integration.
LangChain/LlamaIndex RAG pipelines, Hugging Face Transformer classifiers, XGBoost/LightGBM models, OpenAI & Azure OpenAI for generation.
FAISS, ChromaDB, Pinecone & Weaviate vector stores, embeddings, PostgreSQL for structured data.
Azure ML, AWS (EC2/S3/Lambda/SageMaker), Docker, Kubernetes, CI/CD, MLflow, Weights & Biases, model monitoring & drift detection.
Where I've worked, and the cloud regions I deploy to. Click a pin.
Click a pin on the globe to see details
05 · Experience
06 · Projects
AI document search assistant (Python, LlamaIndex, OpenAI API, ChromaDB, embeddings) retrieving policy, ticket, and knowledge-base answers, improving retrieval accuracy by 31%.
ML classification system (Scikit-learn, LightGBM, Pandas, SQL) predicting recurring issue categories, improving ticket prioritization accuracy by 29%.
Recommendation workflow (NLP preprocessing, sentence embeddings, FAISS similarity search, Streamlit) that cut article lookup time by 30%.
07 · Leadership & Ownership
Shipping AI features end-to-end means owning the problem, not just the notebook.
Partner directly with product, QA, and data teams to define model inputs, acceptance criteria, and validation rules, cutting AI feature rework by up to 25% before release.
Choose the right tool for the problem, whether that's RAG vs. fine-tuning or classical ML vs. deep learning, and defend those calls with evaluation data (ROC-AUC, F1, A/B results).
Build Streamlit and Gradio prototypes so non-technical stakeholders can see model behavior early, shrinking requirement-clarification cycles by up to 26%.
Own failure-case testing and answer validation against source documents with QA before rollout, treating "it sounds right" as insufficient for production.
08 · Certifications
09 · Education
A core engineering discipline built the analytical and systems-thinking foundation I later applied to AI/ML, formalized through focused AI/ML & GenAI certifications and 4+ years of hands-on production work.
10 · Contact
Open to AI/ML Engineer roles and interesting problems in Generative AI, RAG, and applied ML. I'm quick to reply.
srinath.koyi@applywizard.ai
USA
Use the call / message buttons in the corner →