Shail K Patel

AI Engineer. I take LLM and RAG products from zero to production.

Connect Resume

EXPERIENCE

RestaurantPilot.ai (Restaurant Tech Startup)

Founding Machine Learning Engineer

Seattle, USA (Remote)
Nov. 2025 – present
  • Built production RAG with hybrid vector and keyword retrieval and fused reranking over ChromaDB and MongoDB Atlas Vector Search.
  • Own 5 LLM extraction pipelines pairing OCR and PDF parsing with schema-constrained structured output; cut per-invoice inference cost 75% and LLM call latency 59%.
  • Built a tool-calling agent loop over 38 internal tools and a provider-agnostic layer routing across multiple LLM providers; lead a 4-person team across ML, backend, and frontend.

GET MY SPACE (Parking Management Startup)

Machine Learning Engineer

Ahmedabad, India
Jan. 2025 – Jun. 2025
  • Built a real-time ANPR pipeline over live camera feeds, with format-validated fallback across multiple recognition models, improving accuracy 73%.
  • Designed the system’s architecture and deployed computer vision models behind production APIs, cutting recognition latency 34%.

PUBLICATIONS

Source Project PredictGrad →

A Two-Stage, Leakage-Aware Framework for Early Academic Risk Detection in Undergraduate Engineering Cohorts

Student Performance Prediction, Early Warning Systems, Ensemble Learning, Stacked Classification, Leakage Prevention, SHAP
  • Proposed and implemented a leakage-aware, two-stage ML framework to forecast student academic decline
  • Applied leakage-aware modeling to ensure predictions are temporally valid and unbiased
  • Evaluated 56 regression pipelines per subject (224 total) to benchmark academic performance forecasting methods
  • Tested 50 classification pipelines for risk detection; best-performing stacking ensemble achieved recall 0.657
  • Applied explainability (SHAP) to support model interpretability and trust
Paper doi.org/10.5281/zenodo.19686376
1 / 1

PROJECTS

Open to Work

Code →
Python, FastAPI, Qdrant, LiteLLM, LaTeX, Docker
  • Self-hosted app that builds tailored resumes from GitHub evidence
  • Extracts skills from repo manifests and READMEs, weighted by recency
  • Semantic search picks only grounded content, auto-fits LaTeX to page

PredictGrad

Live → Code → Paper →
Ensemble Stacking · BayesSearchCV · Voting Regressor · SHAP/XAI · Boruta Feature Selection · Class Imbalance Handling
  • Built a full ML pipeline to predict future semester marks and detect academic risk
  • Regression: Voting Regressor (Ridge + Lasso + ElasticNet)
  • Classification: Stacking (CatBoost, LGBM, ExtraTrees)

Régions Inégales: Regional Firm-Creation Attribution Model

Live → Code →
Python, XGBoost, SHAP, Pandas, Streamlit
  • Built a decade-long panel from nine official French sources, harmonizing inconsistent schemas across 96 metropolitan departments
  • XGBoost model with leave-one-department-out cross-validation, achieving R² = 0.678 on unseen departments
  • SHAP attribution showed opportunity features (58%) vastly outweighed necessity (20%), with unemployment ranking last

CERTIFICATES

SKILLS

LLM & GenAI

  • LLMs
  • RAG
  • Generative AI
  • Agentic AI
  • Prompt Engineering
  • Structured Extraction
  • Evaluation Datasets
  • LangChain
  • LangGraph
  • LlamaIndex
  • LiteLLM
  • Hugging Face
  • OpenAI & Gemini APIs

Retrieval

  • ChromaDB
  • MongoDB Atlas
  • Embeddings
  • Hybrid Retrieval
  • Reranking

Backend & Infra

  • Python
  • FastAPI
  • Django
  • REST APIs
  • Docker
  • AWS (S3, Lambda)
  • Git
  • Linux
  • Streamlit

Data & ML

  • PostgreSQL
  • MongoDB
  • Redis
  • SQL
  • NoSQL
  • Pandas
  • NumPy
  • PyTorch
  • Scikit-learn
  • NLP
  • Computer Vision
  • MLOps
  • Model Deployment
Get in touch

Let's Work
Together

I'm always excited to build systems that actually matter. If you have a product vision that needs intelligent architecture, or a technical challenge that needs ownership from first principles, let's talk.

Email Me shailpatel.connect@gmail.com
↗
LinkedIn shailkpatel
↗
GitHub shailkpatel
↗