Sid Valecha

Experiences

Sonorus Labs

September 2025 - Present

Founder/Independent Developer

Madison, WI

Developing an AI-powered SaaS creative tool for musicians that transforms natural-language prompts into theory-validated chord progressions and playable MIDI output.
Architected a production-ready system using FastAPI, Next.js, and a modular Python engine for intent parsing, deterministic music generation, and MIDI rendering.
Implemented music-theory–driven generation, MIDI export, browser playback, and Supabase backed persistence to support user history and usage analytics.

73 Strings

June 2024 - August 2024

AI/ML Engineering Intern

New York, New York, United States

Built and deployed LLM-based financial document parsers end-to-end, reducing processing time by 30% and improving extraction accuracy for private-equity analytics.
Developed, trained, and productionized a chart detection and classification model, improving precision to 93% (+12%) and recall to 94.4% (+11%). Achieved 92.2% mAP, and performed hyper-parameter tuning to reduce false positives and improve model reliability in production.

United Nations (OICT)

May 2024 - June 2024

AI/ML and Full Stack Intern

Remote

Built and deployed a Retrieval-Augmented Generation (RAG) system for a UN research portal, enabling context-aware querying over large collections of unstructured documents.
Designed and implemented document ingestion, preprocessing, and embedding pipelines, supporting scalable retrieval and low-latency inference in production.
Implemented research repository file-handling and storage workflows to support AI-driven document processing, contributing beyond core ML responsibilities and learning production systems on the job.

Featured Projects

From early data projects to the tools I’m actively shipping today.

Praetor

In active development

An open-source, local-first orchestration harness for coding agents, built as the trust gate around autonomous execution. It queues scoped tasks, routes them through Claude Code or Codex, verifies outputs with real commands, and enforces adversarial review before work is marked ready. It runs through review cycles and blocks merges until PRs pass all gates.

Agent Orchestration Git Worktrees Adversarial Review Developer Tools

Code

CC Plugin for Codex

In active development

Reverse engineered OpenAI's Claude Code plugin pattern and rebuilt it as an open-sourced project so I can route work between Codex and Claude. I use Claude for planning, design, and adversarial review, and use Codex for implementation across the build flow, with tracked job history and permission-aware handoffs between agents.

Model Orchestration Planning Adversarial Review Open Source Plugin

Code

Ticker Talk: Talk to your stocks!

AI-powered natural language stock analysis app that turns queries like “forecast AAPL for 7 days” into full technical analysis. Computes indicators and ARIMA forecasts in Python, generates charts, and adds AI-generated explanations with guardrails. Built with FastAPI + Next.js and optimized with caching.

Time Series Forecasting ARIMA Natural Language Interface Caching Strategy

Demo Code

Chord Probability Engine

A chord progression probability explorer built with N-gram models. Analyzes chord sequences from the Chordonomicon dataset and provides an interactive web interface for exploring chord transition probabilities using unigram, bigram, and trigram models with interpolation.

Python Machine Learning N-gram Models Markov Chains Web Development

Demo Code