Inspirational Vision Background
The Ideas-to-Life Philosophy

Structured Edge AI Evaluator

Exploring whether specialized micro-evaluation models at the edge can deliver deterministic alignment scoring and conversion bottleneck diagnosis without the latency and cost of frontier LLMs.

shipping
Career DevelopmentProductivityGenerative AI
Structured Edge AI Evaluator

Experiment Details

This experiment follows the Ideas to Life Experiments Charter. Experiments are learning artefacts, not products. They test reusable patterns, not one-off implementations.

Structured Edge AI Evaluator

Why This Exists

Evaluating textual fit—such as candidate CVs against enterprise job specifications or freelance proposals against client briefs—is broken. Heuristic ATS tools miss semantic nuance, while heavy frontier LLMs introduce multi-second latency, variable JSON outputs, and high token costs. Reviewers and applicants need instant, structured, deterministic feedback to identify critical gaps before submission.

What This Experiment Explores

What question(s) are you trying to answer?

  • Can dedicated, lightweight evaluation models (typesafe/jev) on edge runtimes deliver sub-second, reliable rubric scores?
  • Does zero-shot structured JSON evaluation eliminate the parsing drift and prompt instability common in general-purpose conversational models?
  • Can discrete multi-axis scoring and bottleneck isolation improve user iteration velocity across career and proposal workflows?

What Was Built

  • Describe the shape of the artefact or system, not implementation details. Focus on the reusable pattern being tested, not the specific instance.
  • Edge Evaluation Engine: A lightweight Cloudflare Worker service dispatching structured scoring schemas directly to specialized edge models with IP rate-limiting and tier controls.
  • Multi-Axis Alignment Evaluator (CV-to-JD): An interactive tool assessing candidates across 4 core fit dimensions with discrete gap isolation and interview recommendations.
  • Proposal Conversion Diagnostician (Upwork): A structured rubric evaluator analyzing freelance proposals across 6 conversion drivers (hook, comprehension, authority, tone, CTA, budget fit).

Key Trade-offs

What you deliberately chose not to do — and why.

  • Structured Scoring over Long-Form Generative Rewrites: Deliberately avoided generating rewritten CVs or proposals to focus strictly on objective evaluation clarity and avoid hallucinated claims.
  • Specialized Edge Micro-Models over Frontier Reasoning Models: Traded open-ended conversational capabilities for sub-500ms edge execution speed, zero JSON parsing failures, and minimal token overhead.

Current Status

This experiment is currently in the Shipping phase.

Interactive web artifacts and public API endpoints are live with rate limiting and VIP tiers, validating real-world feedback loops across CV fit and proposal scoring. Ensure alignment with the Definition of Done described in the Experiments Charter (pattern validated, documented, and linked to evidence where applicable).

Only include links that exist and matter.