Problem Statement Atlas

A living directory of concrete problem statements across the domains most worth having a real, built project for — the master list Case Studies and this site's actual built projects both plug into. Each entry tracks one status:

Not started — no dedicated content or project yet.
Content only — the concepts are covered somewhere on this site, but there's no dedicated deep-dive case study or a real built project for this specific problem statement yet.
In progress — an actual build has started but isn't finished/published yet; no link until it lands.
Real project built — an actual, deployed, documented system exists for it, linked below.

This page gets updated as case studies and real projects land — it's the index, not the content itself. See Case Studies for the deep-dive write-ups (problem → requirements → scale estimation → architecture → SOTA comparison → failure modes → cost/latency tradeoffs) as they're built out, and The Project Ladder for how a project's rigor scales by level.

Production RAG

Problem statement	Status	Related content
Multi-source enterprise RAG (docs + tickets + chat, unified retrieval)	Content only	RAG
Hybrid search + reranking at production scale	Content only	RAG, Embedding & Reranker Inference
Long-document RAG (legal/medical) with chunking-strategy tradeoffs	Not started	—

Agentic RAG

Problem statement	Status	Related content
Agentic RAG with query planning + multi-hop retrieval	Not started	—
Self-correcting RAG (critique-and-retry on low-confidence retrieval)	Not started	—

Multi-Agent Orchestration

Problem statement	Status	Related content
Supervisor-pattern multi-agent research assistant	Content only	Agent Architectures, Multi-Agent Systems
Multi-agent coding assistant with sandboxed execution	Not started	Sandboxing

Recommendation & Ranking at Scale

Problem statement	Status	Related content
Two-stage retrieval + ranking recommender on a real dataset	Content only	Recommender Systems, Learning to Rank
Real-time personalization with online/incremental learning	Not started	—

Search Systems

Problem statement	Status	Related content
Hybrid lexical + semantic search over a real corpus	Content only	Vector Databases
Learning-to-rank search reranker, benchmarked	Content only	Learning to Rank

LLM Serving & Inference at Scale

Problem statement	Status	Related content
Multi-engine benchmark: vLLM vs. SGLang vs. TensorRT-LLM vs. llama.cpp	Content only	LLM Inference Engines — also a named Level 6 capstone
Autoscaling LLM serving with real cost/latency tradeoffs	Content only	LLM Hosting & Serving Patterns, AI Cost Engineering

Model Routing & Gateways

Problem statement	Status	Related content
Cost-aware LLM router (cheap vs. expensive model by query difficulty)	Content only	AI Cost Engineering
Unified multi-provider LLM gateway with fallback	Not started	—

Fine-Tuning Under Constraints

Problem statement	Status	Related content
LoRA/QLoRA fine-tune on a limited GPU budget	Content only	Training Pipeline — PEFT
Distilling a large model's task performance into a small one	Content only	AI Cost Engineering

MLOps / CI-CD for ML

Problem statement	Status	Related content
Full CI/CD pipeline with automated retraining gate	Content only	CI/CD & ML CI/CD
Production churn model with canary deployment + drift monitoring	Content only	The Full MLOps Architecture — Five Capstone Projects
Statistically-calibrated drift monitoring for a production classifier (empirical false-positive-rate validation against known, injected drift — not assumed correct)	Real project built	`ml-drift-monitor`, Monitoring & Drift — A Real, Proven Example, Case Study — Fraud Detection's Monitoring step

Model Compression & Edge Deployment

Problem statement	Status	Related content
Model compression & edge-deployment benchmark suite (quantization/distillation/pruning compared on real accuracy-vs-latency-vs-size tradeoffs)	In progress	Quantization, LLM Inference Optimization — Quantization

AI Security (Prompt Injection / RAG Poisoning)

Problem statement	Status	Related content
RAG poisoning red-team + defense evaluation	Content only	AI Security
Prompt injection detection/defense pipeline	Content only	AI Security

Agent Evaluation

Problem statement	Status	Related content
Agent trajectory + tool-call accuracy evaluation harness	Content only	LLM/RAG/Agent Evaluation
Automated agent regression testing (jailbreak/tool-call CI gate)	Content only	ML & LLM Testing

Next: The Project Ladder — once you've picked a problem statement above, scope it to the right level of rigor.