Samir Hadiyal — AI/ML Engineer

About Me

Know Who I Am

Background & Focus

I'm Samir Hadiyal, I am learning AI.
I spend most of my time in the space where computer vision, retrieval, and LLMs meet — training models, building pipelines that combine multiple retrieval strategies, and building evaluation systems that check their own judges before trusting them. The part I care about most isn't making a model sound smart, it's figuring out exactly where it shouldn't be trusted, and building deterministic guardrails there instead.
Always happy to talk computer vision, RAG, or AI systems.

Experience

Work Experience

Real-world experience building ML systems and Python applications.

Internship

ML Engineer Intern

Sparks to Ideas

Ahmedabad, Gujarat May 2025 – July 2025

Key Responsibilities

Gained practical experience with NumPy, Pandas, Scikit-learn, and Matplotlib through hands-on ML exercises
Applied data preprocessing, feature engineering, and visualization techniques on sample datasets
Implemented and evaluated introductory machine learning models under mentor guidance

Key Achievements

Built and evaluated multiple ML models across supervised learning exercises under direct mentorship
Strengthened end-to-end ML workflow skills from raw data to trained model in a real internship setting

Internship

Python Intern

Ramora Technologies

Rajkot, Gujarat June 2024 – June 2024

Key Responsibilities

Learned core Python programming concepts including loops, functions, file handling, and OOP
Applied Python fundamentals by solving supervised coding assignments and debugging exercises
Built beginner-level Python programs following structured development practices

Key Achievements

Completed a full month-long internship building foundational Python proficiency from first principles
Delivered multiple structured coding assignments following professional development practices

Education

My Education

Diploma in Computer Engineering

A.V. Parekh Technical Institute, Rajkot

Full-time CGPA: 8.67

Key Highlights

Specialization in AI/ML with hands-on coursework spanning deep learning, computer vision, and NLP

Built a strong foundation in Python, Java, SQL, and core data structures alongside applied ML coursework

Maintained an 8.67 CGPA while independently building and shipping production-style AI projects

Projects

Featured Projects

End-to-end AI applications — from research to production.

LLM-Eval-CI/CD

Multi-document RAG service where the eval system is the point, hybrid retrieval + reranking, LLM-judge scoring calibrated against human ground truth, and a CI/CD gate that blocks regressions, not just low scores.

View on GitHub

Key Outcomes

1 Identified and solved a common RAG blind spot: built a judge-calibration harness to verify an LLM evaluator agrees with human-labeled ground truth before trusting it as a CI/CD gate.

2 Engineered a hybrid retrieval pipeline (ChromaDB dense + BM25 sparse, fused via Reciprocal Rank Fusion) with FlashRank cross-encoder reranking to maximize context precision before generation.

3 Built a regression-aware GitHub Actions quality gate failing a PR if scores drop >5% below a stored production baseline validated across a 35-case benchmark spanning factual retrieval, multi-chunk synthesis, and guardrail cases.

4 App: https://llm-eval-cicd.streamlit.app

RAG LangChain Groq FastAPI GitHub Actions ChromaDB

SnakeSense

A multimodal snake species identification and clinical triage system combining computer vision, retrieval-augmented generation, and a deterministic rule-based safety engine to help identify medically significant Indian snake species and guide urgency response.

View on GitHub

Key Outcomes

1 Trained an EfficientNet-B0 classifier on 15 medically significant Indian snake species, achieving 87% top-3 test accuracy

2 Achieved sub-100ms GPU inference served through a FastAPI backend

3 Built a ChromaDB + Groq retrieval system delivering trilingual (English/Hindi/Gujarati) responses grounded in WHO/NCDC clinical guidelines

4 Designed an LLM-independent, rule-based triage engine for auditable, reproducible urgency classification

RAG PyTorch FastAPI LLMs ChromaDB

AI Powered Dialogue Summarizer

An end-to-end abstractive dialogue summarization tool fine-tuned on real conversational data, served through a FastAPI REST API with an interactive web interface.

View on GitHub

Key Outcomes

1 Fine-tuned a T5 Transformer model for abstractive summarization using the SAMSum dataset

2 Developed a FastAPI REST API with an interactive web interface for user-submitted text

3 Implemented text preprocessing, tokenization, and beam search decoding for quality output

4 Built GPU/CPU device selection logic for efficient inference across environments

Python Hugging Face Transformers PyTorch FastAPI

Review Authenticity & Sentiment Analysis

An end-to-end machine learning pipeline that detects potentially fake reviews and performs sentiment analysis on genuine reviews at scale.

View on GitHub

Key Outcomes

1 Designed and implemented a full ML pipeline for fake review detection and sentiment analysis