Senior Backend AI Systems Engineer, Python / NLP / Knowledge Graphs

Well Inked Memoirs is building technology to help preserve real human stories. We combine interviews, human editors, and AI-assisted systems to create thoughtful memoirs and biographies. We're now looking for a senior Python backend engineer to help build the internal systems that turn long interview transcripts into structured, reviewable memory.

This is not a generic chatbot or AI-agent role.

The core challenge is turning long interview transcripts into structured, reviewable memory: speakers, transcript segments, people, events, claims, relationships, dates, uncertainty, and provenance. That structured memory is reviewed by humans and then assembled into chapter-specific context packets for biography generation.

We need someone who can think clearly about messy data, source traceability, human review workflows, and production-quality backend systems.

What You'll Work On

You may work across:

• Python backend services and APIs
• transcript ingestion and processing workflows
• speaker-aware transcript structures
• structured LLM extraction pipelines
• people, event, claim, date, and relationship extraction
• provenance back to transcript segments
• confidence and uncertainty handling
• graph-style data modelling
• human review and correction workflows
• internal editorial and admin tools
• async jobs, queues, workers, and orchestration
• validation, testing, logging, and production hardening

What We're Looking For

Strong candidates will have experience with:

• Python backend development
• FastAPI, Django, Flask, or similar
• database design and data modelling
• Postgres or similar relational databases
• APIs, services, and backend workflows
• async jobs, queues, workers, or orchestration
• structured LLM outputs, JSON schemas, or validation workflows
• production debugging, reliability, and maintainability

Experience with any of the following is highly relevant:

• NLP
• RAG
• entity extraction
• document intelligence
• transcript processing
• diarization or speech-to-text workflows
• Whisper or similar transcription systems
• knowledge graphs or graph databases
• Neo4j or relationship modelling
• evaluation workflows for AI outputs
• human-in-the-loop review systems

What We Are Not Looking For

We are not looking for someone who has only built simple AI wrappers, chatbot demos, or prompt-based prototypes.

We need someone who can reason through backend architecture, edge cases, data quality, source traceability, and long-term maintainability.

AI tools in development are completely fine. What matters is engineering judgment. You should be able to understand, verify, and own the systems you build.

Good Fit

You may be a good fit if you can:

• work from rough technical direction and make it implementable
• ask good questions about system boundaries and failure modes
• design clear backend data models
• keep LLM outputs constrained, validated, and auditable
• think carefully about uncertainty, provenance, and human review
• build pragmatic systems without over-engineering
• communicate clearly in English
• collaborate with a small team on a complex early-stage product

Contract Details

• Remote contract role
• Potential for ongoing work if there is a strong fit
• Hourly or scoped contract arrangements considered

To Apply

Please send to [Confidential Information]:

• a short intro
• your relevant Python/backend experience
• examples of AI, NLP, RAG, transcript, document-processing, or data extraction systems you have built
• your hourly rate or preferred contract structure
• your availability
• links to GitHub, portfolio, CV, LinkedIn, Upwork, or relevant work

Please also briefly answer this question:

Imagine we have a 90-minute interview transcript with imperfect speaker labels, vague dates, repeated names, contradictions, and emotional reflections mixed with factual events. How would you design a backend pipeline to turn this into structured memory with provenance, human review, and chapter-specific context packets

We are especially interested in how you would think about data models, extraction schemas, failure points, uncertainty, and preventing silent errors.