A chatbot is not an AI strategy.
Strategy starts with a user problem worth solving and an honest call on whether AI is the cheapest reliable way to solve it. The interface comes last.
AI Product & Platform Leadership
Shubham
Chandra
I build AI products that make it to production — and stay there.
Thirteen years of data and AI systems at scale: from the first user problem, to the architecture review, to the metrics dashboard that proves it worked.
13 years across
- Morgan Stanley
- Warner Bros. Discovery
- Fox
- EPAM Systems
Executive profile
Most AI features die between the demo and the roadmap.
Not because the model was weak — because nobody designed for retrieval quality, evaluation, cost, latency, and adoption from day one. Building for that gap, turning AI from a prototype into a product capability, is my job.
Over thirteen years I've built data and AI systems at scale for Morgan Stanley, Warner Bros. Discovery, Fox, and EPAM Systems — leading cross-functional teams from 0→1 launches through 1→n hardening. Much of that production work lives behind NDAs, so this site shows my thinking, my toolkit, and what I build in the open.
I hold a Master's in Data Science from the University at Buffalo, SUNY, and a Bachelor's in Computer Science Engineering. The PMP on my wall and the evaluation pipelines in my repos are the same instinct: outcomes you can audit, not demos you can only applaud.
- 13yData & AI at scale
- 4Enterprise employers
- 5Professional certifications
- 0→1to 1→n delivery
How I think about AI products
Positions I'll defend in any architecture review.
Data quality is the AI strategy.
Retrieval, grounding, and evaluation are only as good as the data platform underneath. Fix the pipeline before you tune the prompt.
Evaluation before scale.
If you can't measure whether the system got better, you can't ship it twice. Evals, guardrails, and cost budgets are designed in — never bolted on.
The demo-to-production gap is organizational.
Closing it takes product direction, platform architecture, and change management moving together. That intersection is where I work.
Where I spend my time
Three jobs, one operating loop.
Product direction
Decide what deserves to exist.
User problems worth solving with AI, ruthless prioritization, and honest calls on what ships versus what stays in the lab. Business goals first; model choices follow.
- Use-case discovery
- AI product strategy
- Roadmaps
- Stakeholder alignment
Platform architecture
Design systems that survive contact with users.
GenAI platforms on Bedrock and Snowflake Cortex, RAG and agent architectures, data platforms, evaluation pipelines, and the cost and latency budgets that keep them honest.
- RAG & agents
- Evaluation pipelines
- Data platforms
- Cost & latency budgets
Ship & scale
Lead teams from launch to durability.
Cross-functional delivery from 0→1 launches through 1→n hardening — reliability, observability, adoption, and the metrics that prove the product earned its roadmap slot.
- Cross-functional leadership
- PMP-grade execution
- Observability
- Adoption metrics
Now building
Two live builds, one quality bar.
In active build
Enterprise AI Assistant Platform
An AWS-native conversational AI product over organizational knowledge — Bedrock AgentCore for agent orchestration, Snowflake Cortex for warehouse-native intelligence, RAG over Confluence and S3 with a managed vector store, and deployment trade-offs benchmarked across ECS Fargate, EKS, and Amplify.
Evaluation, guardrails, and cost controls designed in from day one — not bolted on.
- AWS Bedrock
- Snowflake Cortex
- RAG
- Agents
In active build
Computational Astronomy Engine
A from-scratch astronomical computation and 3D visualization product: Swiss Ephemeris for planetary-grade positional accuracy, two rendered sky modes — symbolic chart view and true local-sky 3D projection — and a structured, audit-trailed interpretation pipeline where every output traces to a finding ID.
Same bar I hold production AI to: deterministic inputs, explainable outputs.
- Python
- Swiss Ephemeris
- 3D visualization
- Audit trails
Projects in the open
Built in public, shipped as products.
Open-source work is where I test the same standards I hold enterprise systems to — each of these went from notebook to something a person can actually use.
HireHorizon
GenAIA generative-AI resume optimizer: upload a resume, paste a job description, get alignment feedback, missing keywords, and a profile summary. Live on Hugging Face.
- Python
- LLMs
- Hugging Face
MovieSelect
RecommendersPersonalized movie recommendations over IMDB and TMDB data — recommender logic taken from notebook to a usable Streamlit product with rating, era, and revenue filters.
- Python
- Streamlit
- Recommender systems
Incometric
MLIncome prediction from demographic and socio-economic signals — the segmentation layer behind pricing, credit, and targeting products, deployed as a web app and API.
- Python
- ML pipelines
- Model evaluation
LeafCare
Computer visionPlant-disease diagnosis from leaf images with a CNN trained on PlantVillage — computer vision shipped as a decision tool on Hugging Face Spaces, not a benchmark score.
- CNN
- Image classification
- Streamlit
HomeScope
MLEnd-to-end California housing price prediction — raw data to an interactive app a buyer can actually reason with, with full deployment guidance.
- Random Forest
- Streamlit
- End-to-end
ClarifyMeet-AI
GenAIAn AI meeting companion in active development — the newest of the open builds, applying the same evaluation-first discipline to conversational workflows.
- Python
- LLMs
- In development
Toolkit
The stack behind the systems.
AI & GenAI
- Claude & Claude Code
- AWS Bedrock
- OpenAI
- Azure AI
- RAG
- AI agents
- LLM evaluation
- LangChain
- Hugging Face
Cloud & data platforms
- AWS
- Azure
- GCP
- Snowflake + Cortex
- Databricks
- Apache Spark
- Airflow
- Kafka
- dbt
- Docker
- Kubernetes
- GitHub Actions
ML & analytics
- Python
- PyTorch
- TensorFlow
- scikit-learn
- SQL
- PostgreSQL
- MongoDB
- Tableau
- Power BI
- Streamlit
Credentials
Certified where it counts, educated where it started.
Education
M.S. Data Science · B.E. Computer Science
University at Buffalo, SUNY — New York, USA.
PMI
Project Management Professional
Program execution & delivery.
AWS
Solutions Architect — Professional
Cloud architecture at scale.
Anthropic
Claude Certified Architect — Foundations
Building with Claude & agentic systems.
AWS
Generative AI Developer — Professional
Production GenAI systems on AWS.
Snowflake
SnowPro Core
Cloud data platform & warehousing.
Contact
Building an AI product that has to survive production?
Retrieval quality, evaluation, agent reliability, the demo-to-production gap — I'm always up for that conversation, whether you're a recruiter, a founder, or a product team mid-flight.