Optimize AI-Driven Academic Writing Assistant

We're building an AI writing assistant for students, think Jenni AI + Paperpal. The core pipeline works, but we need a backend/AI specialist to improve accuracy, optimize performance, and prepare it for scale before public launch.


------------------------------------------------------------------------------------------------------------------------------------

Tech Stack:

Backend: FastAPI + async SQLAlchemy 2.0, Python codebase
Data Layer: PostgreSQL, Redis, MinIO/S3
AI Infrastructure: LiteLLM gateway (multi-LLM routing), RAG pipelines

------------------------------------------------------------------------------------------------------------------------------------
What You'll Do:
1-Audit and integrate new evidence sources: Map the gaps in OpenAlex coverage, evaluate 2-3 alternative academic databases, and integrate at least one new source into the retrieval pipeline
2-Build hallucination detection: Implement citation requirement logic so every claim must be traceable to a retrieved document. Add a rejection gate that flags unsupported statements before generation completes. Test and measure citation coverage improvement
3-Fix tone/style drift: Build a coherence checker that scans new paragraphs against the document's existing sections for voice, argument structure, and terminology consistency. Flag mismatches for revision
4-Speed up streaming: Profile the current pipeline for bottlenecks, reduce LLM latency (optimize batching, caching, routing), and improve streaming UI responsiveness. Target: faster first-token output
5-Improve evidence ranking: Refine RAG retrieval scoring so the top results are actually the most relevant. Test reranking models or implement custom scoring logic
Strengthen quality gates: Add checks in the evaluation framework to catch: missing citations, tone inconsistency, unsupported claims. Adjust thresholds and trigger auto-revision appropriately

And more will be discussed in DM
------------------------------------------------------------------------------------------------------------------------------------
You Need:

Production LLM pipeline experience (shipped, not experimental)
1- RAG architecture expertise
2-Python, FastAPI, PostgreSQL, Redis
3-Work with LLMs as infrastructure (APIs)
4-Concrete metrics from previous work (latency, relevance, cost improvements)
5-Know what's possible with existing LLM APIs vs. custom engineering
6-Portfolio: Share GitHub profile\ repo, previous project links, or case study showing your work. We need to see real code

------------------------------------------------------------------------------------------------------------------------------------
Nice to Have:
LiteLLM or multi-provider LLM routing
Vector database tuning
Academic/research writing tools experience (Jenni, Paperpal, similar)
EdTech background
FastAPI at scale