Dieses Stellenangebot ist nicht mehr verfügbar
Über
The context
Asset-Spec is a SaaS for cataloging equipment. We already have a stable equipment entry schema with over 1,000 uploaded items, but the current AI/enrichment backend is unreliable. We need a production-grade AI service that can be consumed by multiple frontends. So we are envisioning a separation of front/backend to derisk and speed up development cylces.
What we need (outcome-focused): A standalone backend service that enriches an existing equipment entry and returns structured results, confidence, pricing rationale, evidence. Critical is strong observability so I can evaluate and control implementation quality based on results without debating tech stack details.
Core capabilities of the AI service:
• Image recognition to Brand/Model (candidates + confidence)
• Brand/Model to Manual/Datasheet retrieval (source URLs + confidence)
• Manual to Structured specs (typed fields + evidence references)
• Pricing: price range by condition/age plus a short rationale and confidence;
Non-negotiables (control & reliability)
• Async job processing: submit - job_id - status - results (for bulk processing via batch JSON; Excel is low priority)
• Observability: correlation IDs (tenant_id, equipment_id, job_id), structured logs, step-level status, stored job outputs, and replayability for failed runs
• Gated writes: AI outputs must be returned as proposals (patches) with confidence/evidence; no silent overwrites of canonical data.
• API contract: versioned REST + OpenAPI
• Dockerized, deployable independently (ideally via github); basic tests (unit + integration)
Technical expectations
We are not prescribing the full stack — we want a developer who can define and implement the appropriate and coste effective architecture for this problem. That said, a few things are fixed:
• The service must be built in Python or Both are acceptable; tell us which you'd choose and why.
• GPT-Vision API will be used for image recognition. If you have a better solution please propose and explain why.
• A Google Search integration (via Apify, or equivalent) is required for manual retrieval. How you handle document (usually pdf) parsing, spec extraction, and pricing lookups is part of what we want to see in your proposal.
• The API must be versioned, fully documented (OpenAPI/Swagger), and designed for consumption by multiple independent frontends.
• The service must be Dockerized and deployable independently from any frontend.
• Data privacy is a consideration — equipment data belongs to our clients and should not be unnecessarily exposed to third-party services. We expect this to factor into your architecture proposal.
• PDF and document parsing capability is required for manual processing — the service must be able to extract structured data from datasheets and technical documents, not just retrieve them.
• Async job queue is required for bulk processing. The service must support background workers, a queue mechanism, and a job status endpoint. Redis-based queues are preferred but open to alternatives.
• Proper logging, error handling, and confidence-based flagging are required — not optional.
Milestone structure (fixed deliverables based on acceptance test
Milestone 0 — Architecture + Contracts (paid, short)
• Deliver:
o I) under 2 page design (components, job model, data flow, privacy), OpenAPI draft, output schemas for proposals/evidence/pricing rationale, logging/metrics plan.
o Ii) proposed tech stack for implementation
• Acceptance: clear API + schemas; test plan; replay/debug story.
Milestone 1 — Job system + Observability baseline
• Deliver: async job API + queue/worker + status endpoints; correlation IDs; structured logging; minimal "hello pipeline" stub.
• Acceptance: I can submit a job, observe step status, retrieve logs by job_id, and replay a failed job deterministically.
Milestone 2 — Recognition + Manual retrieval (with evidence)
• Deliver: image recognition endpoint and manual search step, both independently callable + pipeline wrapper.
• Acceptance: run against my provided test dataset; outputs conform to schema and include confidence + evidence URLs.
Milestone 3 — Spec extraction + Pricing with rationale
• Deliver: spec extraction into structured fields + pricing range + rationale + confidence; modular pricing component.
• Acceptance: run against my provided dataset; schema-valid outputs; pricing includes rationale; failure cases are explainable via logs and status.
Milestone 4 — Hardening + Documentation
• Deliver: tests, CI-ready setup, Docker, OpenAPI final, minimal runbook (env vars, deploy, replay).
• Acceptance: I can deploy from repo, run test suite, and execute the acceptance dataset end-to-end using an API directly from the existing code - base
Who we are looking for
We need a backend specialist who has built AI-integrated services before — not someone learning on the job. The ideal candidate is comfortable designing API contracts, working with LLM APIs, handling async pipelines for bulk processing, and delivering something they would be comfortable putting their name on.
Experience in manufacturing, industrial equipment, or asset management is a bonus but not required. Clear communication and the ability to flag problems early matter more than domain knowledge.
How we'll work
I will provide acceptance datasets (images, PDFs, equipment entries) to validate quality.
Milestones are paid only when the acceptance test passes and outputs match the contract.
Scope & timeline
This is a fixed-scope project. We estimate 4–5 weeks for a complete, tested, and documented implementation. We prefer milestone-based payments tied to working deliverables at each stage. After Milestone 0 we prefer fixed price implementation workpackages
To apply, please include
1. A brief description of a similar project you have built — specifically backend API development with AI or LLM integration.
2. Your stack choice (Python or ) and a short explanation of why.
3. How you would approach the architecture for this service — a few sentences is enough, we are not asking for a full spec.
4. Your timeline estimate and rate.
Generic proposals without relevant experience described will not be considered.
Contract duration of 1 to 3 months.
Mandatory skills: Python, , RESTful API, API Integration
Sprachkenntnisse
- English
Hinweis für Nutzer
Dieses Stellenangebot wurde von einem unserer Partner veröffentlicht. Sie können das Originalangebot einsehen hier.