Fluffless: Deterministic AI Proposal Engine Project

Fluffless: Deterministic AI Proposal Engine

Project Overview / TL;DR

A high-performance Next.js web application that forces LLMs to write hyper-direct, senior-level technical pitches by anchoring generation to a persistent, client-side profile vault.

The Problem:

The freelance market is flooded with generic, AI-generated proposals that read like bloated cover letters. Clients ignore them because they lack technical depth and focus on "marketing fluff" rather than execution. Additionally, existing tools create massive UX friction by forcing users to manually re-type their past work experience for every single pitch.

The Solution:

I architected and deployed Fluffless to solve both the output quality and the user friction. I bypassed standard AI wrappers to build a constraints-based engine that only generates technically sound, highly specific proposals.

Key Technical Execution:

Client-Side "Profile Vault": To eliminate repetitive data entry, I engineered a persistent LocalStorage architecture. Users input their raw technical resume once, and the application silently passes this base context to the AI for all future generations, reducing user input to just the target job description.

Heavyweight AI & Prompt Engineering: I integrated the Groq API utilizing the massive Llama-3.3-70B reasoning model. I engineered a strict, one-shot prompt architecture loaded with negative constraints to completely strip out default "AI-speak" (e.g., words like leverage, delve, seamless) and force a rigid 4-step psychological sales structure (Hook -> Proof -> Execution -> Technical CTA).

Native Streaming Architecture: When the standard Vercel AI SDK introduced silent failures due to protocol mismatches, I bypassed the abstraction entirely. I built a custom, native JavaScript fetch and TextDecoder implementation to guarantee 100% stable, low-latency text streaming directly to the UI.

My Role:

Lead Full-Stack Developer & Prompt Engineer

Tools & Technologies:

Next.js (App Router)

React

Tailwind CSS

Groq API (Llama-3.3-70B)

Vercel

Project Links:

Live Site: https://fluffless.vercel.app/