Samit Fernandes

Building software that reasons with data and communicates through language.

The Narrative

I believe the future of software lies beyond forms and dashboards, where conversations become the primary way people interact with intelligent systems. My work focuses on integrating AI into software so that applications can understand user intent, reason over structured data, and provide meaningful assistance.

Rather than treating AI as a standalone feature, I enjoy building systems where natural language, backend engineering, and intelligent reasoning work together seamlessly. I am driven by the challenge of creating software that is conversational, explainable, and grounded in reliable engineering principles.

Technical Foundations

Programming

Python
JavaScript
SQL
Git

Language Intelligence

Tokenization & Embeddings
Attention Mechanisms
Representation Learning
Information Retrieval
Probabilistic Models
Model Fine-tuning & Evaluation

Systems & Data

Data Modeling
PostgreSQL
Vector Retrieval
Retrieval-Augmented Generation
Distributed Data Processing

Research & Engineering

An index of experimental systems, research notebooks, and engineering explorations documenting work in natural language processing, retrieval, and intelligent software.

01 · Language Representation

Embedding Visualizer
Prototype

Interactive 3D visualization of semantic vector spaces using dimensionality reduction.

Vector Math Data Viz
Contextual Vectors
Notebook

Analyzing polysemy and dynamic word representations derived from attention heads.

Semantics Attention
Contrastive Maps
Experiment

Aligning multilingual semantic spaces through unsupervised contrastive learning techniques.

Algorithms Mapping

02 · Language Understanding

BabyLLM
Prototype

An experimental language learning system exploring conversational interactions and memory.

Modeling Memory
ResumAI
System

Retrieval-augmented document understanding and semantic chunking for contextual generation.

Chunking Pipeline
Sentiment Analysis
Experiment

Multilingual text classification and linguistic feature extraction pipeline.

Classification Extraction

03 · Information Retrieval

Nyaya
System

Semantic search and document intelligence system for navigating complex legal text.

Indexing Search Engine
BM25
Notebook

Custom lexical retrieval engine built from standard ranking functions and term frequencies.

Lexical Scoring
Dense Retrieval
Experiment

Vector-based semantic search exploring embedding similarity metrics and ANN architectures.

Embeddings Distances
RAG
System

End-to-end retrieval augmented generation pipeline for factual question answering.

Retrieval Generation

04 · Foundation Models

BPE Tokenizer
Notebook

Subword tokenization algorithm and vocabulary construction implemented from scratch.

Parsing Encoding
Transformer
Notebook

First-principles implementation of attention mechanisms and positional encoding.

Architecture Matrices
Mini GPT
Prototype

Lightweight autoregressive language model built to understand sequence prediction.

Training Inference
Token Optimizer
Experiment

Experimental techniques targeting vocabulary efficiency and structural compression.

Optimization Data Structures

05 · Reasoning Systems

Knowledge Graph QA
Prototype

Query parsing and path traversal for graph-based question answering systems.

Graphs Traversal
Neural-Symbolic AI
Experiment

Experimental framework combining deep neural networks with strict logic rules.

Logic Heuristics
Logic Engine
Notebook

First-principles implementation of a forward and backward chaining inference engine.

Inference Parsing

06 · Conversational Software

FinzFox
System

Conversational financial intelligence merging structured data with natural language.

Analytics UI / UX
SalesMate
Prototype

Interactive analytical tool transforming business data into conversational summaries.

Dashboards Interfaces

Knowledge Hub

An index of active research logs, developmental roadmaps, and conceptual essays documenting theoretical explorations and practical implementations.