Aditya Thoomati

// projects

Tritone Analytics · New York / Portland

Royalty Gap Analysis Pipeline

Built a data pipeline to identify unpaid music royalties by cross-referencing mechanical licensing data, Spotify stream estimates, and actual artist payouts. Processed hundreds of gigabytes of data to surface compensation gaps across entire artist catalogues — uncovering roughly $40,000/year in unpaid royalties owed to artists and writers missing from royalty chains. Also built a chat interface so the team could query the dataset and generate charts on demand.

Clickhouse Python Pandas Data Engineering AWS Bedrock Chat Interface

Data Actions · Startup

AI Analytics Agent System

Built an AI-powered analytics system with multiple agents and custom FastMCP tools over a semantic data layer. Implemented tool use / function calling, ReAct reasoning loops, multi-agent orchestration, structured JSON outputs, and long-term agent memory backed by a vector store. Users query business metrics in plain language and get answers with charts; a graph-based root cause analysis agent — built on a decision context graph — pinpoints which dimension or lever is driving a given anomaly.

FastAPI FastMCP LangGraph Agno LangChain Langfuse MCP Servers Python

Freelance · Hyderabad

Salon — Data & Web Overhaul

Analysed CRM data from Fresha to identify high-performing customer segments and seasonal trends. Rebuilt the website from scratch with strong SEO, premium content, and targeted CTAs — including a WhatsApp offer flow that converted site visitors into direct enquiries. Polished the Google Business profile to improve local discovery. Result: daily revenue is up, enquiry quality improved, and the owner is now planning to open a second location within six months.

Python Pandas HTML/CSS SEO GitHub Pages Google Ads

IIT Madras · MTech AI & ML

Fine-tuning LLMs, Vision & Audio Models

Research and implementation work on fine-tuning large language models, vision models, and audio models as part of the MTech programme at IIT Madras. Applied LoRA and QLoRA for parameter-efficient fine-tuning, and worked with seq2seq architectures alongside the Hugging Face ecosystem. Also studied LLM internals end-to-end — tokenisation, attention mechanisms, residual streams, and how fine-tuning actually changes model behaviour rather than just running the scripts.

LoRA / QLoRA Seq2Seq Hugging Face PyTorch Python

Internship · AI Consultant · Feb – Sep 2023

RAG Systems & AI POC Development

Built multiple proof-of-concept AI systems during a 7-month internship with an independent AI consultant. Projects included document Q&A pipelines, knowledge base chatbots, and natural language data retrieval agents — each built on RAG architectures with vector search, embedding pipelines, and LLM orchestration. Deployed on GCP.

RAG LangChain LlamaIndex ChromaDB Qdrant pgvector GCP Python

Freelance · Multiple Clients

RAG Chatbots for Business Workflows

Built RAG-powered chatbots for clients including Waxwing.ai — ingesting internal knowledge bases into vector stores, then exposing them as chat interfaces to automate workflows and answer domain-specific queries.

LangChain OpenAI API ChromaDB RAG Python

Freelance · Spudnik Farms

E-Commerce Site — Laravel

Picked up a stalled custom Laravel e-commerce site and got it across the line with another developer. Inherited an incomplete codebase, identified what was left, and shipped a working storefront for the client.

Laravel PHP E-Commerce

IIT Madras · App Dev

Kanban App — Full Stack

Full-stack Kanban board with a Vue.js frontend, Flask backend, REST APIs, and automated daily report emails via background jobs.

Vue.js Flask REST APIs Celery

IIT Madras · ML

Consumer Purchase Pattern Prediction

Predicted consumer purchase behaviour from an e-commerce dataset — feature engineering, model selection, and evaluation.

Python Pandas Scikit-learn NumPy

// about

Developer with a builder's instinct

I'm a developer and AI engineer currently pursuing an MTech in AI & ML at IIT Madras, where I also completed my B.S. in Data Science and Programming with a CGPA of 8.19.

My background is unusual. Before going deep on tech, I spent three years as a Growth Marketer at Chargebee — growing the inbound lead funnel by 180% through SEO, shipping growth tools, and running campaigns that got product people to rethink their build-vs-buy decisions. That context shapes how I think about what I build: I'm not just writing code, I'm solving problems that matter to a business.

Alongside the tech work, I help run a small family agro-tech venture focused on fruit production and food processing for export — which is where a lot of my interest in applying data and AI to real-world operations comes from.

I take on freelance projects and am open to full-time roles where I can work on data engineering, AI agents, or applied ML — especially in early-stage or product-focused teams.

// ai & agents

LangChain LangGraph LlamaIndex Agno FastMCP MCP Servers RAG Langfuse LLM Evals Prompt Engineering

// models & llms

Anthropic API Gemini API OpenAI API AWS Bedrock Ollama Hugging Face Sentence Transformers Cohere Embed LoRA / QLoRA Transformer Internals PyTorch Scikit-learn

// data & databases

Clickhouse PostgreSQL MySQL MongoDB ChromaDB Qdrant pgvector Pandas NumPy SQL dbt Kafka Redis

// cloud & devops

GCP Vertex AI Cloud Run Cloud Storage AWS S3 AWS Lambda Docker GitHub Actions pytest

// web & tools

Python FastAPI Flask Vue.js Chart.js Streamlit Plotly Matplotlib REST APIs asyncio Celery BeautifulSoup Jupyter n8n