Gulshan Kumar | AI Engineer | M.Sc. Artificial Intelligence

01. Core Academic & Professional Credentials

Master of ScienceGrade: 101 / 110 (91.8%)

Master in Artificial Intelligence

University of Verona (Università degli Studi di Verona), Italy

Graduation: 21 July 2026 • Fully Funded Merit Scholarship

Thesis:"Personalized User Modeling for Personalized Query Recommendation" (Advisor: Department of Computer Science).

Current PositionApril 2026 – Present

Artificial Intelligence Engineer

Fintech Scaleup • Verona, Italy

Engineered 300x query speedup (154s → 0.5s) via deterministic SQL routing.
Architected on-prem local LLM RAG pipelines with Ollama and FAISS.
Modernized legacy MS Access / PHP routines to async Polars/FastAPI ETL.

02. Verified Performance Metrics & Benchmarks

300x Speedup

Fintech SQL Query Routing

154s → 0.5s Latency

101 / 110

Verona M.Sc. in AI

Academic Distinction

< 18 ms P95

FAISS Vector Search

Bipartite Node2Vec

86% Accuracy

Cholec80 Surgical AI

ResNet-50 + LSTM

0.042 RMSE

Macro GDP Forecaster

40+ Yr IMF Dataset

0.89 NDCG@5

Thesis Retrieval Metric

MS MARCO Benchmark

130K+ Reach

Educational Community

500+ Mentored

7 Languages

Native Localization

English C1, Italian B1

03. Flagship Systems & Production Case Studies

Enterprise Fintech Analytics & Local LLM RAG Pipeline

300x latency reduction, zero cloud API token cost, strict data sovereignty.

Stack: Python, FastAPI, Ollama, Phi-3, FAISS, MySQL, Redis, Docker

View Study

Master’s Thesis: Personalized Query Recommendation via Graph Embeddings

Bipartite Node2Vec random walk embeddings yielding 0.89 NDCG@5 and <18ms retrieval.

Stack: PyTorch, FAISS IndexFlatIP, NetworkX, Scikit-learn, SentenceTransformers

View Study

Multimodal Surgical Workflow Analysis (Cholec80 Laparoscopy Benchmark)

86% temporal phase recognition accuracy for continuous laparoscopic procedures.

Stack: PyTorch, ResNet-50, Bi-LSTM, OpenCV, Albumentations, CUDA

View Study

Multi-Country Macroeconomic GDP Forecasting Platform

Ensemble model (ARIMA + XGBoost + Stacked LSTM) serving 20+ policy analysts.

Stack: Python, Flask, Stacked LSTM, Optuna, Polars, Power BI, World Bank API

View Study

04. Technical & Engineering Taxonomy

AI / ML & LLMs:

PyTorch, SentenceTransformers, Scikit-learn, XGBoost, Ollama, Phi-3, Llama-3, FAISS, ChromaDB, RAG, BM25, GGUF Quantization, Node2Vec Graph Embeddings.

Backend & Cloud:

FastAPI, Django, Python 3.11, Polars, Pandas, Redis, MySQL, PostgreSQL, Docker, Linux, CI/CD, Sliding-Window Rate Limiting, Async Event Loops.

Spoken Languages:

English (C1 Fluent / Professional), Italian (B1 Working / Verona), Urdu (Native), Hindi (Native), Sindhi (Native).

01. Core Academic & Professional Credentials

Master of ScienceGrade: 101 / 110 (91.8%)

Master in Artificial Intelligence

University of Verona (Università degli Studi di Verona), Italy

Graduation: 21 July 2026 • Fully Funded Merit Scholarship

Thesis:"Personalized User Modeling for Personalized Query Recommendation" (Advisor: Department of Computer Science).

Current PositionApril 2026 – Present

Artificial Intelligence Engineer

Fintech Scaleup • Verona, Italy

Engineered 300x query speedup (154s → 0.5s) via deterministic SQL routing.
Architected on-prem local LLM RAG pipelines with Ollama and FAISS.
Modernized legacy MS Access / PHP routines to async Polars/FastAPI ETL.

02. Verified Performance Metrics & Benchmarks

300x Speedup

Fintech SQL Query Routing

154s → 0.5s Latency

101 / 110

Verona M.Sc. in AI

Academic Distinction

< 18 ms P95

FAISS Vector Search

Bipartite Node2Vec

86% Accuracy

Cholec80 Surgical AI

ResNet-50 + LSTM

0.042 RMSE

Macro GDP Forecaster

40+ Yr IMF Dataset

0.89 NDCG@5

Thesis Retrieval Metric

MS MARCO Benchmark

130K+ Reach

Educational Community

500+ Mentored

7 Languages

Native Localization

English C1, Italian B1

03. Flagship Systems & Production Case Studies

Enterprise Fintech Analytics & Local LLM RAG Pipeline

300x latency reduction, zero cloud API token cost, strict data sovereignty.

Stack: Python, FastAPI, Ollama, Phi-3, FAISS, MySQL, Redis, Docker

View Study

Master’s Thesis: Personalized Query Recommendation via Graph Embeddings

Bipartite Node2Vec random walk embeddings yielding 0.89 NDCG@5 and <18ms retrieval.

Stack: PyTorch, FAISS IndexFlatIP, NetworkX, Scikit-learn, SentenceTransformers

View Study

Multimodal Surgical Workflow Analysis (Cholec80 Laparoscopy Benchmark)

86% temporal phase recognition accuracy for continuous laparoscopic procedures.

Stack: PyTorch, ResNet-50, Bi-LSTM, OpenCV, Albumentations, CUDA

View Study

Multi-Country Macroeconomic GDP Forecasting Platform

Ensemble model (ARIMA + XGBoost + Stacked LSTM) serving 20+ policy analysts.

Stack: Python, Flask, Stacked LSTM, Optuna, Polars, Power BI, World Bank API

View Study

04. Technical & Engineering Taxonomy

AI / ML & LLMs:

PyTorch, SentenceTransformers, Scikit-learn, XGBoost, Ollama, Phi-3, Llama-3, FAISS, ChromaDB, RAG, BM25, GGUF Quantization, Node2Vec Graph Embeddings.

Backend & Cloud:

FastAPI, Django, Python 3.11, Polars, Pandas, Redis, MySQL, PostgreSQL, Docker, Linux, CI/CD, Sliding-Window Rate Limiting, Async Event Loops.

Spoken Languages:

English (C1 Fluent / Professional), Italian (B1 Working / Verona), Urdu (Native), Hindi (Native), Sindhi (Native).