contradictory-body / Strata

What is Strata?

Strata is a Python-based multimodal job search agent that acts as your personal career coach, analyzing resumes, job descriptions, and screenshots while remembering your preferences like target roles or salary expectations across sessions. It handles long conversations without context overflow using hierarchical memory—short-term chat history, compressed summaries, and long-term semantic storage—plus tools for JD breakdowns and web searches via Tavily. Run it via CLI with commands like /file for uploading PDFs/images or /profile to check your persistent job profile.

Why is it gaining traction?

In a sea of flat-memory agents, Strata stands out with its cognitive-inspired three-layer memory that cuts token use by over 50% and boosts recall accuracy, letting you iterate on job strategies over dozens of turns without losing key details. Multimodal support shines for real-world job hunting—parse Word docs, scan resume images with vision LLMs, and get structured feedback—backed by ablation experiments showing clear wins over baselines. Developers digging github multimodal ai or multimodal rag pipelines appreciate the drop-in ReAct loop with async memory writes.

Who should use this?

AI builders prototyping multimodal job agents or github multimodal llm setups for career tools. Job hunters in tech needing a persistent assistant for JD matching, interview prep, or preference tracking during extended hunts. Folks exploring multimodal ai jobs, like those in cfl multimodal jobs in luxembourg, who want file uploads and search without rebuilding context each time.

Verdict

Worth forking for niche multimodal job workflows or as a hierarchical memory benchmark—24 stars and solid docs/experiments signal early promise despite 1.0% credibility. Too alpha for production (light tests, single-maintainer), but clone and tweak the CLI for your next agent hack.

(198 words)