Sunil Samson Suresh
Senior Data Scientist
I build GenAI, agentic systems and production ML that actually ships, end to end on Azure and Databricks.
- Based in
- Hyderabad, India
- Experience
- 5+ years
- Domains
- Insurance · Immigration tech · Legal
// 5+ yrs · hyderabad, india
I build GenAI, agentic systems and production ML that actually ships, end to end on Azure and Databricks.
currently: [TODO: what I'm building / learning]
// about
I build AI that actually ships.
Senior Data Scientist with 5+ years building and deploying production-grade LLM and ML systems across insurance, immigration tech and legal, specialising in GenAI, agentic systems and end-to-end delivery on Azure and Databricks.
I lead technical decisions across cross-functional teams, mentor junior members, and set engineering standards for how AI features are scoped, built and shipped.
My foundation: Python, PyTorch, TensorFlow, SQL, Docker, Kubernetes and FastAPI, backed by MLOps practices, open-source contributions to LlamaIndex and an IEEE publication.
Hyderabad, India
- 5+ yrs Building production AI & ML systems Insurance · Immigration tech · Legal
- Azure & Databricks End-to-end ML delivery · Databricks Certified ML Associate
- MLOps stack FastAPI · Docker · Kubernetes · MLflow · CI/CD
- GenAI & agents LangGraph · LlamaIndex · RAG · Transformers
// experience
Five years, four teams, real-world stakes.
From crop-risk forecasting and legal NLP to agentic GenAI for immigration cases and fraud & severity models for insurance claims.
-
Now
Senior Data Scientist
Hyderabad, IndiaInsurance · Claims analytics
Leading fraud and severity modelling for Allied World's Claims Analytics team, and turning manual claims workflows into automated ML pipelines.
What I do
- Lead fraud and severity data modelling for Allied World's Claims Analytics team across 9,000 Program Business Claims, building signals and features with claims stakeholders to flag high-risk claims early so adjusters can intervene before costs escalate.
- Primary technical liaison between data science and business stakeholders: I evaluate technical feasibility in discovery sessions, present findings in plain language and tie model outputs to concrete business decisions.
- Lead the end-to-end transformation of document triage, severity prioritisation and intervention decisioning, replacing manual workflows with automated ML pipelines for measurable gains in throughput, accuracy and cost.
Focus
Fraud & severity modelsFeature engineeringDocument triageML pipelinesBusiness liaison
-
Data Scientist
Tallinn, EstoniaImmigration tech · GenAI
Set the technical vision for platform data science and GenAI, and shipped agentic LLM systems as production microservices.
What I did
- Led the technical vision for platform data science and GenAI integration, partnering with Product and Engineering to set engineering standards for scoping, building and shipping AI-powered features.
- Architected end-to-end autonomous decision workflows using GPT, Mistral and Claude on LangGraph and LlamaIndex with open-source models, deployed as production-grade FastAPI microservices powering real-time automation at scale.
- Designed a conversational RAG-based agentic system for the message composer, backed by MongoDB vector search, grounding LLM outputs in live case data for accurate, status-specific responses with no stale outputs.
- Built OCR and computer-vision pipelines to parse identity documents across multiple countries, using vision models to extract and validate structured PII fields.
Stack
LangGraphLlamaIndexGPTMistralClaudeFastAPIMongoDB vector searchRAGOCRComputer vision
-
Jr. Data Scientist
Hyderabad, IndiaLegal · NLP
Built legal-domain NLP, from fine-tuned transformers to extraction pipelines, plus Bayesian fraud-risk forecasting.
What I did
- Fine-tuned transformer models including BERT, RoBERTa and LegalBERT for multi-class NLP tasks (entity recognition, clause categorisation, document labelling), optimising accuracy across diverse legal corpora through systematic experimentation.
- Built end-to-end NLP pipelines for structured information extraction from large legal document corpora, covering entity tagging and liability clause identification, directly powering downstream compliance and decision systems for enterprise clients.
- Applied Bayesian probabilistic modelling with Pyro for fraud-risk forecasting and uncertainty quantification, producing calibrated confidence estimates so stakeholders could make risk-based decisions with explicit uncertainty bounds.
Stack
BERTRoBERTaLegalBERTNLP pipelinesPyroBayesian modelling
-
Data Scientist Intern
Hyderabad, IndiaAgriculture · Forecasting
Forecast crop lifecycle events, pest outbreaks and disease risk from weather, satellite and temporal signals.
What I did
- Developed time-series forecasting models to predict crop lifecycle events, pest outbreaks and disease risks from historical and environmental data.
- Built predictive analytics pipelines integrating weather, satellite imagery and temporal signals for early-warning systems, with feature engineering on temporal and geospatial data to improve forecasting accuracy.
Focus
Time-series forecastingSatellite imageryGeospatial featuresEarly-warning pipelines
Every model looks good in a notebook. I build the part where it ships.
// skill map
A production toolkit, tuned for GenAI & ML.
Every skill on my resume. Hover or tap one to see where it shows up in my experience.
genai
GenAI & agents
ml
Machine learning
cloud + data
Cloud & data
mlops
MLOps & engineering
where_used(skill) # pick a skill above
// contact
Let's build something that ships.
Open to conversations about GenAI, agentic systems and production ML. Email is the fastest way to reach me.
Direct line
Talk to me directly.
The quickest way to reach me is a short email about what you're working on.
-
email
loading… -
based_in
Hyderabad, India
- linkedinin/sunilsamson (opens in a new tab)
- githubgithub.com/sunilsamsonsuresh (opens in a new tab)
- resumeDownload CV (PDF)