Building intelligent systems across enterprise AI, personalisation, and language understanding.
JUL 2025 — PRESENT01
S&P Global
Senior Data Scientist
Building ASTRA, an enterprise agent platform processing 23B+ tokens a month, and leading WhoBot, a knowledge platform serving 30K+ employees.
More about the work +
Built ASTRA on AWS AgentCore over proprietary enterprise data, serving 5K+ employees and 16K+ AI sessions.
Built an evaluation harness with 400+ golden traces across tool use and retrieval, reducing the escape rate from 23% to under 5%.
Own ASTRA’s guardrails: PII redaction, prompt-injection detection, and full trace audit logging. The platform cleared InfoSec review for restricted enterprise data.
Led WhoBot to answer ownership, risk, and product questions across siloed internal systems. Built its multi-agent knowledge graph across eight source systems, extracting 2M+ entity relationships.
Used an LLM-judge validation loop to raise knowledge-graph edge precision from 71% to 89%.
Selected as one of 15 firm-wide Vision Professors. Authored an agent-design and fine-tuning curriculum now used to onboard 150+ AI teams.
AI agents · Knowledge graphs · Evaluation & guardrails
OCT 2023 — JUL 202502
Wipro
AI Lead, GenAI Practice
Led 15 GenAI engineers delivering enterprise AI and agentic systems. Launched root-cause analysis solutions that reduced infrastructure mean time to recovery by 15%.
More about the work +
Fine-tuned and deployed domain-specific small language models on secure, GPU-based on-premises infrastructure.
Partnered with client stakeholders across 6+ enterprise engagements, from solution scoping through production deployment.
Defined architecture, evaluation criteria, and rollout plans for enterprise AI and infrastructure diagnostic solutions.
Generative AI · Fine-tuning · Model deployment
MAY 2021 — OCT 202303
Let's Cooee
Founding Engineer and Data Scientist
Built and scaled core AI and data products as an early founding engineer, including real-time purchase-intent systems processing 11M+ e-commerce sessions a day.
More about the work +
Designed an event-driven architecture that improved data-processing efficiency by more than 60%.
Developed an LLM-powered analytics system using tool calling over PostgreSQL, contributing to a 4%+ conversion uplift.
Automated model training, evaluation, deployment, monitoring, and retraining pipelines.
Personalisation · Real-time systems · MLOps
JUN 2019 — MAY 202104
Riverus
Associate Data Scientist
Built machine learning systems for large-scale legal corpora, achieving 90% precision in extracting high-value analytics.
More about the work +
Improved fuzzy text-matching performance by 25% for quoted-text comparisons across legal datasets.
Developed and deployed an ensemble of transformer and supervised learning models supporting one-class classification for 160K+ users.
Transformers · NLP · Information extraction
02 / PRACTICE
Curiosity meets engineering.
My work connects machine learning research with the systems that make it useful.
Intelligence
Generative AI, LLMs, NLP, deep learning, reinforcement learning, graph networks