SiddharthThakar

AutoDS — autonomous data scientist

A full-stack AutoML application that takes a raw table or a time series and returns a finished analysis: exploratory profiling, preprocessing, model training, diagnostics, explainability, a written report, and an inference endpoint. Built so an analyst can reach a defensible model without hand-writing the same forty notebook cells again.

Stack

Python, Streamlit, FastAPI, PostgreSQL, Docker

Covers

Tabular ML and time-series forecasting

Role

Design, build and deployment

Canada Labour & Affordability Intelligence

An analytics platform that joins Job Bank, Statistics Canada and CMHC data so you can compare Canadian cities on the things that decide where you move: which occupations are hiring, what they pay, what rent does to that wage, and where the trend is heading. Includes forecasts and automated data-quality checks that flag a source before it quietly breaks a report.

Stack

Power BI, DAX, PostgreSQL, Python

Sources

Job Bank, Statistics Canada, CMHC

Role

Data model, ETL, report design

Ask a PDF, get a cited answer

A retrieval-augmented question answering system for documents that runs entirely on your own machine — Ollama embeddings, ChromaDB retrieval, a re-ranking pass to cut weak matches, and Llama 3.2:3b for generation. No document ever leaves the laptop, which is the point.

Stack

Python, Streamlit, Ollama, ChromaDB, Llama 3.2:3b

Focus

Retrieval quality and re-ranking

Role

Solo build

Extractor.io

Commodity trading back offices receive broker contract notes as PDFs, every broker in a different layout, and someone re-keys thousands of individual fills into Excel to check the bill. Extractor.io detects the format, pulls every fill with its timestamp, runs FIFO matching for realised P&L, values open lots against MCX settlement prices, and rebuilds the day's charges so the broker's own bill can be compared line by line.

Stack

FastAPI, SQLAlchemy, PostgreSQL, Camelot, pdfplumber, OpenCV, React

My part

Extraction and reconciliation reporting

Status

Active development

InvoiceOS

Accounting firms do the work and then struggle to get paid, because chasing an invoice is awkward and always the lowest-priority task on the desk. InvoiceOS raises GST-compliant invoices, works out the tax split from the client's location, and then chases payment on a timed follow-up ladder that runs from a courtesy nudge to a formal notice. Fintale reports a 32% reduction in 60+ day receivables and 91% of invoices paid before escalation.

Stack

React, Vite, Tailwind, Express, SQLite, node-cron

My part

Invoice generation and receivables reporting

Status

Live in production

CFO Intelligence

Accountants spend four to eight hours per client every month rebuilding the same Excel management report. This reads a Trial Balance exported from any accounting package and returns a full dashboard — 20+ KPIs, ratios, cash forecasts, risk alerts — benchmarked against live government-sourced industry data, then exports it as a branded board pack. Fintale puts the turnaround at about a minute per client.

Stack

React, Recharts, Express, PostgreSQL, Python

My part

Dashboards, KPI logic and reporting

Status

Live in production