|
|
|
📌 AI Systems Engineer & Open Source Builder specializing in Interpretable ML, Vector Databases (Qdrant/Chroma), and Custom AI Agents for global startups & research teams.
"Models should not only predict well — they should explain well."
I approach modeling through three core principles:
- Statistical validity before scale — Ensuring assumptions and tests are mathematically grounded before scaling computation.
- Interpretability before optimization — Understanding feature importance and logic paths before squeezing marginal decimal gains.
- Domain meaning before deployment — Ensuring features translate directly to clinical or business reality.
My research interests include:
- Permutation-based, resampling, and nonparametric inference
- Interpretable and explainable machine learning (post-hoc & intrinsic)
- Dimensionality reduction with geometric and statistical intuition
- Robustness, stability, and noise-aware modeling
- Translating statistical theory into clinically actionable insights
| Category | Technologies |
|---|---|
| 🧠 Modeling & AI | Python, R, PyTorch, TensorFlow, scikit-learn |
| 🔧 Infrastructure & Pipelines | Docker, PostgreSQL, AWS, GCP, Git, CI/CD Workflows |
| 📦 Multi-Language Dev | TypeScript / Node.js, Java (Maven), C# (.NET), Ruby |
  
Integrates arXiv and Semantic Scholar directly into AI IDEs with page-level PDF extraction and citation graph traversal. |
 
Nonparametric Combination (NPC) and bootstrap-based clinical risk stratification model for rare-disease clinical research. |
  
Clinical NLP platform for unstructured record mining, entity extraction, clinical sentiment analysis, and automated ICD coding. |
  
Scalable clustering framework (KMeans++, DBSCAN, BIRCH, OPTICS) applied to NYC Taxi mobility (12M+) and fraud detection. |
  
General-purpose statistical library for Nonparametric Combination of permutation tests and multi-criteria severity ranking. |
  
Containerized cross-engine vector database synchronization tool to migrate and replicate embeddings between Chroma and Qdrant. |
I actively contribute to major AI/ML open-source projects with bug fixes, performance improvements, and core infrastructure enhancements:
| Repository | PR | Description | Status |
|---|---|---|---|
| qdrant/qdrant | #1264 | Vector search engine improvement | ✅ Merged |
| run-llama/llama_index | #22343 | MinioReader basename collision fix | 🔍 Under Review |
| chroma-core/chroma | #7432 | Embedding search improvement | 🔍 Under Review |
| logspace-ai/langflow | #14051 | Workflow engine enhancement | 🔍 Under Review |
| lancedb/lancedb | #3661 | Retrieval pipeline fix | 🏁 Closed |
| milvus-io/pymilvus | #3686 | Python SDK improvement | 🔍 Under Review |
| explodinggradients/ragas | #2850 | Evaluation framework fix | 🔍 Under Review |
| cleanlab/cleanlab | #1321 | Data-centric AI enhancement | 🔍 Under Review |
| public-apis/public-apis | #6592 | Reported 5 broken API links | 📋 Issue Filed |
-
langchain-ai/langchain#39018 — Blob Byte Stream Serialization Fix (Under Review 🔍)
- System Impact: Fixed a
NotImplementedErrorinBlob.as_bytes_io()when initialized with in-memory string data, enabling seamless byte-stream decoding across document loaders and RAG data pipelines. - Key Technologies: Python, LangChain Core, In-Memory Streams, Data Pipelines.
- System Impact: Fixed a
-
ray-project/ray#64864 — Core Python & RLlib Docstring Refactoring (Approved ✅)
- System Impact: Resolved multiple typographical errors in core signature inspection and RLlib model catalogs, passing full CI check suites.
- Key Technologies: Python, Ray RLlib, Distributed Computing.
-
dmlc/xgboost#12335 — XGBRanker Score Validation Check (Under Review 🔍)
- System Impact: Enforced query ID (
qid) validation inXGBRanker.score()to prevent silent fallback to single-query NDCG evaluation in ranking models. - Key Technologies: Python, C++ Backend, XGBoost Scikit-learn API, Ranking Metrics.
- System Impact: Enforced query ID (
-
lancedb/lancedb#3661 — Workspace Manifest & Dependency Refactoring (Approved ✅)
- System Impact: Resolved a blocking issue in the workspace linter (
cargo deny) triggered by yanked upstream crates (likespin v0.10.0), and resolved duplicate developer dependency declarations. - Key Technologies: Rust, Cargo Workspace, CI/CD Linters.
- System Impact: Resolved a blocking issue in the workspace linter (
-
milvus-io/pymilvus#3686 — Environment Configuration Security & Type Safety (Under Review 🔍)
- System Impact: Patched SDK settings parser to prevent unauthorized loading of
.envconfigurations when thePYTHON_DOTENV_DISABLEDflag is set, and resolved a runtimeTypeErrorcaused by uninitialized variables. - Key Technologies: Python, python-dotenv, Environment Management.
- System Impact: Patched SDK settings parser to prevent unauthorized loading of
C# / .NET AI Trading Agent & Portfolio Strategy Validator using SPRT, Sharpe/Sortino Ratios |
Ruby AI Financial Fraud & Anomalous Transaction Auditor with consensus verification |
Java Demographic Fairness Credit Risk Evaluator with bias metrics |
Permutation-Based Analysis of Clinical Variables in Necrotizing Fasciitis Using NPC and Bootstrap
Mathematics, MDPI (2025)
This work introduces a permutation-based, nonparametric framework for analyzing clinical variables in necrotizing fasciitis. By combining Nonparametric Combination (NPC) methodology with bootstrap techniques, the study enables robust inference under small-sample and distribution-free conditions, with an emphasis on interpretability and clinical relevance.
The study demonstrates how permutation-based inference can outperform classical parametric approaches in rare-disease clinical settings.
🔗 https://www.mdpi.com/2227-7390/13/17/2869
- 🧬 High-Precision Biomedical AI: Designing mathematically rigorous, permutation-based nonparametric statistical engines for rare diseases and small-sample clinical datasets.
- 📈 Robust & Explainable ML (XAI): Developing invariant feature attribution and post-hoc explanation frameworks that maintain mathematical consistency under severe covariate and distribution shifts.
- 🛡️ Production AI Guardrails & Validation: Building automated evaluation suites, stability tests, and performance diagnostics to guarantee safety, reproducibility, and alignment in high-stakes clinical and financial ML deployments.
- 🔧 Developer Tooling & Agentic Integrations: Architecting FastMCP (Model Context Protocol) servers to seamlessly link scientific databases (arXiv, PubMed, gnomAD) with next-generation LLM agents and coding systems.
|
|
|
|
|
|
