Tahsin Reza
Builder of Local AI Systems

CEO of Sunset Pulse. Architect of the TAH (Take a Hint) standard — giving AI instant domain expertise without the cost of massive context windows.

Visit SunsetPulse.app GitHub TAH Research
About

Local-first. Domain-expert. Efficient by design.

Tahsin Reza is a software engineer and AI systems builder based in Texas, and the CEO and founder of Sunset Pulse. His work centers on a core conviction: AI should be fast, affordable, and genuinely knowledgeable about its domain — without requiring massive context windows or expensive API calls to large models.

Outside the codebase, Tahsin also operates the cafe at Sunset Gas and Grill, his family's business.

He is the creator of the TAH (Take a Hint) standard — a binary knowledge cartridge format that encodes domain expertise locally, using probabilistic Bloom filters for sub-millisecond rejection of non-matching content and byte-offset seeking to land directly on the answer. By his own published measurements, the result is dramatic: querying a computer architecture textbook costs under two cents in tokens instead of over five dollars.

SunsetWars defines the TAH knowledge architecture; SunsetPulse is its flagship production application — see Projects below for details.

“Rather than routing all tasks through a single large language model, direct specialized commands to dedicated AI workers. Local knowledge + smaller models + structured cartridges beats one massive model with a huge context window.”

Philosophy

Local-first intelligence. Domain expertise encoded in cartridges, not brute-forced through context. Smaller, cheaper models + structured knowledge outperform larger models on specialized tasks.

Currently Building

Alongside SunsetPulse (see Projects, below), Tahsin is building RentCastJamie (RentCast API integration) and ReRasterizer, and is seeking agents and developers to pilot and contribute.

Projects

Open Source Work

Real Estate AI

SunsetPulse

SunsetPulse's Atlas Globe -- a 3D spatial map of bound real-estate knowledge cartridges plotted as glowing colored points across a dark, gridded globe

A living atlas for North Texas real estate intelligence, place memory, and market context — built with Next.js 14, Supabase, Three.js, and Ollama (Gemma-2). Operated under SunsetCollective. Features Jamie (AI reasoning engine for risk assessment and neighborhood context), a 3D spatial viewer with Atlas Globe Hub, walkable Virtual World Hub, and Unified Property Stage, yield-tracking market analytics, and the Neural Vibe Engine (VibeSimulacrum) — an adaptive styling layer that uses Jamie’s linguistic analysis to extract deterministic interface vibe signatures and morph the platform’s visual presentation consistently across all components — all powered by private TAH knowledge cartridges.

Version 0.3.0 (August 2026) adds a Research Desk — accepts unstructured lead text, verifies facts asynchronously, and routes leads into the correct pipeline stage automatically. The platform now supports multi-tenant branded agent sub-sites, with a Vibe CMS for managing branding through draft, revision, and publish workflows.

github.com/tahsrz/sunset-pulse →
TAH Standard

SunsetWars / TAH

The TAH (Take a Hint) AI gateway — a high-performance knowledge cartridge system that gives LLMs instant domain expertise. Probabilistic Bloom filters for sub-millisecond rejection of non-matching content, byte-offset seeking for precise retrieval, a self-contained binary cartridge (.tah) carrying its own Bloom filter, shard index, and payload — with an optional Memoria “Atlas Handshake” packaging that splits the index into a companion .hat — plus provenance linking to source URLs and ingestion timestamps. Connects to Ollama, Groq, OpenRouter, and Telegram. Python 88%, with C# and TypeScript integrations.

github.com/tahsrz/SunsetWars →

TAH Research: We have conducted extensive independent research on the TAH standard and cartridge format, including architectural deep dives, adversarial reviews, ground-truth benchmarks, and integration work. Read the full research compendium →