Principal ML Engineer @ Databricks | Author | IEEE Senior Member
I build production-grade ML and GenAI systems, with a focus on deployment, evaluation, and reliability. My contributions include implementing core evaluation capabilities in MLflow and building tooling used in enterprise AI systems.
MLflow is one of the most widely adopted open-source ML lifecycle platforms globally (23K+ stars, 18M+ monthly PyPI downloads).
Implemented 3 of 5 third-party scorer integrations for MLflow's GenAI evaluation framework. Each contribution was independently reviewed, approved, and merged by senior MLflow maintainers:
| Integration | PR | Maintainer Review | Production Impact |
|---|---|---|---|
| Phoenix (Arize) | #19473 | Reviewed by @smoorjani, @B-Step62 | Hallucination detection, relevance scoring |
| TruLens | #19492 | Reviewed by @smoorjani, @B-Step62 | Groundedness, context relevance, agent evaluation |
| Guardrails AI | #20038 | Reviewed by @smoorjani | Safety validators (toxicity, PII, jailbreak detection) |
These capabilities are now available to ML practitioners and enterprises worldwide.
| PR | Capability | Status | Impact |
|---|---|---|---|
| #19152 | LLM Judge inference parameters | ✅ Merged | Temperature, top_p control for evaluation |
| #19248 | Configurable scorer parallelism | ✅ Merged | MLFLOW_GENAI_EVAL_MAX_SCORER_WORKERS |
| #20344 | UV package manager integration | 🔄 In Review | Automatic dependency inference |
Bug fixes independently reviewed and merged by external maintainers:
| Repository | PR | Issue Resolved |
|---|---|---|
| truera/trulens | #2328 | Instrumentation crash on non-callable objects |
| truera/trulens | #2308 | Databricks structured outputs compatibility |
Open contributions under review:
| Repository | PR | Proposed Fix |
|---|---|---|
| langchain-ai/langgraph | #6547 | Type signature for conditional edges |
| langchain-ai/langgraph | #6544 | functools.partial handling in ToolNode |
Production deployment capability enabling MLflow models to run on serverless GPU infrastructure:
pip install mlflow-modal-deploy
mlflow deployments create -t modal -m models:/my-model/1 --name my-deployment- Auto-scaling from zero to thousands of GPUs (T4 → H200)
- Sub-second cold starts
- Native MLflow deployment interface
Practical Machine Learning on Databricks
Packt Publishing, 2023 | 244 pages
End-to-end guide for building production ML systems. Best-seller in category.
| Paper | Focus | Venue |
|---|---|---|
| The Semantic Illusion | Hallucination detection failure modes | arXiv:2512.15068 |
| Demystifying Large Language Models | LLM architecture survey | IJCET |
| Reinforcement Learning for Real-World Impact | RL applications | IJSRCET |
| AI in Healthcare | Clinical ML pipelines | IRJMETS |
| Event | Topic |
|---|---|
| TechFutures 2025 (NYC) | End-to-End MLOps Workshop |
| Data Con LA 2022 | Databricks Feature Store |
| Data Con LA 2021 | Fraud Detection at Scale |
| NYU Guest Lecture | ML Pipelines with Apache Spark |
IEEE Senior Member — Recognition for significant contributions to the profession (requires 10+ years experience and documented achievements)
