____  ____   ___      _ _____ ____ _____      _  _____ _        _    ____
   |  _ \|  _ \ / _ \    | | ____/ ___|_   _|    / \|_   _| |      / \  / ___|
   | |_) | |_) | | | |_  | |  _|| |     | |     / _ \ | | | |     / _ \ \___ \
   |  __/|  _ <| |_| | |_| | |__| |___  | |    / ___ \| | | |___ / ___ \ ___) |
   |_|   |_| \_\\___/ \___/|_____\____| |_|   /_/   \_\_| |_____/_/   \_\____/

   121 projects  *  16 tracks  *  superintelligence alignment  *  quantum  *  1 nation-scale master

Project Atlas

121 build-ready project briefs for an edge-AI, security and data engineer, each one sized, budgeted, phased and gated. Seventy-one are level-5: fifty masterpieces, twenty at the frontier of AI alignment and quantum computing, and one sovereign AI blueprint for a nation of 100 million.

This repo is a map, not an implementation. Every brief is a plan for a project that has not been built yet. Nothing here claims a measured result. What it does claim is that the plans are complete and consistent, and that claim is machine-checked: the briefs are generated from tracks/*.yaml, and CI fails if the pages, the index, the counts or the cross-references drift apart.

   by the numbers            counted from tracks/*.yaml by tools/validate.py
     briefs = 121    tracks = 16    level-5 = 71
     masterpiece tier = 50    frontier tier = 20    grand master = 1
     every brief: 4 phases, 2+ gate metrics, 2+ defend questions, an ethics line

   verify it yourself        don't take the prose on trust
     pip install pyyaml pytest
     pytest -q                        # schema and consistency checks over every brief
     python tools/validate.py         # fields, sections, links, counts, upgrade paths
     python tools/render.py --check    # the pages still match the data that generated them

The flagship project is LANTERN, an offline flood search-and-rescue kit. The atlas is the map of what comes next. Each brief has the mission, the architecture, the kit, a phased campaign, a measurable gate, the ethics line, a resume bullet waiting for real numbers, and the questions you must be able to answer out loud.

The compass

                                 [ H ] FRONTIER
                                 47-50 research
                                       ^
          [ G ] PEOPLE & SPORT         |          [ A ] EDGE AI & ROBOTICS
          42-46                        |          01-08
                         \             |             /
                          \            |            /
   [ F ] VISION & GEO ------------- ( YOU ) ------------- [ B ] GENERATIVE AI
   36-41                          /    |    \             09-15
                                 /     |     \
                                /      |      \
          [ E ] CLOUD & MLOPS          |          [ C ] AI SAFETY & GOVERNANCE
          29-35                        v          16-21
                                 [ D ] CYBER DEFENSE
                                 22-28

How to pick the next one

   1. Does it reuse something you already built?      (LANTERN rover, map pack, Guardian ...)
   2. Does it fill a gap a job posting asked for?      (check 3 real postings)
   3. Can you show a measured number in 8 weeks?       (if not, pick a lower level)
   4. Would you explain it at dinner without slides?   (if not, it's not your story yet)

Rule of the atlas: one active project at a time next to LANTERN. A project gets its own repo only on the day you start it, and goes public when it has a first measured result.

Levels

   [#....]  1  a weekend-scale build, one new skill
   [##...]  2  a few months, a clear pipeline, public data
   [###..]  3  real hardware or real evaluation work, a few moving parts
   [####.]  4  research-flavoured, distributed or safety-critical; plan for setbacks
   [#####]  5  reserved for LANTERN-scale capstones

The masterpiece tier (51-100)

                      __________________________________
                     |   M A S T E R P I E C E   T I E R |
                     |__________________________________|
                                      |
       +-------------+-------------+--+----------+-------------+
       |             |             |             |             |
   [ I ] 51-60   [ J ] 61-70   [ K ] 71-80   [ L ] 81-90   [ M ] 91-100
   AUTONOMOUS    FOUNDATION    TRUST &       PLANET &      CIVIC, SCIENCE
   SYSTEMS       MODELS &      SECURITY &    CLIMATE &     & HUMAN
                 AI INFRA      SOVEREIGNTY   INFRA         PERFORMANCE

Every masterpiece brief adds a system map, a technology-readiness target, a publication venue, deliverables, and the standards and regulations it must meet. Each runs 12–24 months.

Every original project (01–50) now carries a masterpiece upgrade path (from upgrades.yaml) showing how it grows into one or more level-5 builds. The professional bar for all of them is the Masterpiece Standard: twelve pillars from requirements to proof of impact.

   the honest math:  50 masterpieces x 12-24 months  =  a career, not a year.
   pick ONE masterpiece next to LANTERN. finish it. publish it. then pick the next.

The frontier tier (101-120) and the grand master (121)

                    *            .        [ 121 NATIONSTACK ]        .          *
                          .            sovereign AI for 100M people
                                     /                            \
           [ N ] 101-110  SUPERINTELLIGENCE ALIGNMENT       [ O ] 111-120  QUANTUM
           interpretability, AI control, sandbagging,      error-correction decoders, honest QML,
           model organisms, scalable oversight, evals,     error mitigation, compilers, QKD,
           deception probes, reward hacking, CoT, compute  GPU simulators, verified quantum agents,
           governance                                       resource estimates for nations

No one builds superintelligence in a garage. The frontier labs and AI safety institutes are hiring people who can make increasingly capable systems understandable, controllable and measurable, and that's what track N trains. Track O is honest quantum: beat the classical baseline or publish that it didn't.

Every frontier page has a Who sees this section naming the fellowship, institute, hackathon or open-source project where that work gets in front of leaders. The Visibility Playbook turns them into a 12-month plan.

NATIONSTACK (121) assembles the whole atlas into a sovereign AI ecosystem for a country of 100 million people: ten pillars, a reference architecture, a working miniature, and a capacity model that turns population into GPUs, megawatts and euros with every assumption visible. Start at master/nationstack/.

The atlas

Track A: Edge AI and Robotics

Intelligence where the cloud can't reach.

#	Codename	Project	Level	Months
01	TIDEWATCH	River-gauge early warning on solar sensor nodes	##...	4
02	HIVEMIND	Beehive health monitor with on-device audio AI	##...	3
03	FORESTER	Wildfire smoke spotter for the Palatinate Forest	###..	5
04	BRIDGEKEEPER	Concrete crack inspection with an edge vision pipeline	###..	4
05	VINEGUARD	Vineyard disease detection for Pfalz winemakers	##...	4
06	QUIETNET	Cloud-free voice control for the home	#....	2
07	PACKMULE	Follow-me cargo rover for volunteers	###..	3
08	ICEROAD	Road ice prediction from low-cost roadside sensors	##...	3

Track B: Generative AI and LLM Systems

Language models that earn their answers.

#	Codename	Project	Level	Months
09	AMTSDEUTSCH	Offline explainer for German government letters	##...	3
10	VERITAS	Citation-checked RAG for technical manuals	###..	3
11	SKILLBRIDGE-COPILOT	Military-to-civilian skills translator	##...	3
12	LOOM	Distil a big model's skill into a tiny one	####.	3
13	ECHO	Offline meeting notes on a laptop	##...	2
14	PATCHWORK	Code-review agent that writes the missing tests	###..	3
15	GRANITE	Regression-test harness for LLM applications	##...	2

Track C: AI Safety and Governance

Trust is a feature you can test.

#	Codename	Project	Level	Months
16	TRIPWIRE	Prompt-injection firewall for LLM apps	###..	3
17	LEDGER	Tamper-evident audit log service for AI decisions	##...	2
18	REDCELL	Automated red-team harness for AI agents	###..	3
19	CHARTER	AI Act documentation generator	##...	2
20	FAIRLINE	Bias audit toolkit for tabular models	##...	2
21	SANDBOX	Capability-limited sandbox for agent tools	####.	3

Track D: Cyber Defense

Defend the network you actually have.

#	Codename	Project	Level	Months
22	SENTRY	Home-network intrusion detection with ML anomaly scoring	##...	3
23	HONEYCOMB	Honeypot fleet with LLM-written threat summaries	##...	2
24	PHISHNET	Explainable phishing classifier	##...	3
25	SBOM-RADAR	Software supply-chain risk dashboard	##...	2
26	ZEROTRUST-LAB	Zero-trust access lab	###..	2
27	LOGLENS	SIEM-lite with LLM-assisted triage	###..	3
28	GHOSTPRINT	Firmware integrity checker for IoT devices	###..	3

Track E: Cloud, Data and MLOps

Models are easy. Production is the project.

#	Codename	Project	Level	Months
29	CONVEYOR	End-to-end MLOps platform on k3s	####.	4
30	RIVERFLOW	Streaming lakehouse for German open water data	###..	3
31	FEATUREFORGE	Minimal feature store built from scratch	###..	3
32	COLDSTART	Serverless inference cost optimiser	###..	2
33	WATCHTOWER	Model drift and data-quality monitoring	##...	2
34	DATACOMPASS	Data lineage and catalogue	##...	2
35	SPOTFLEET	Fault-tolerant GPU training on spot instances	###..	2

Track F: Computer Vision and Geospatial

See the world from the ground and from orbit.

#	Codename	Project	Level	Months
36	ORBITAL	Sentinel-2 change detection for floods and burn scars	###..	3
37	ROOFTOP	Solar potential map from open elevation data	###..	3
38	TRAFFICLENS	Privacy-first traffic counting at the edge	###..	3
39	GREENLINE	Urban tree canopy and heat map	##...	2
40	POTHOLE-PATROL	Dashcam road damage detection	##...	2
41	HERITAGE3D	3D scans of Palatinate castle ruins	###..	2

Track G: People, Sport and Community

Tech that helps someone you can name.

#	Codename	Project	Level	Months
42	HURDLE	Hurdle technique analysis with pose estimation	##...	2
43	POOLSHARK	Pool shot analysis and training coach	##...	2
44	CLIMB-TRANSITION	Veteran transition dashboard	##...	3
45	RECEIPTLENS	Grocery receipt scanner and price tracker	#....	2
46	SPRECHEN	Offline German speaking coach	##...	3

Track H: Frontier and Research

Where the next job titles come from.

#	Codename	Project	Level	Months
47	NEUROSPARK	Spiking neural networks for ultra-low-power sensing	####.	3
48	QUANTUMSAFE	Post-quantum crypto migration for IoT	####.	3
49	SWARM	Multi-robot search coordination in simulation	####.	4
50	HEARTH	Home energy digital twin with reinforcement learning	####.	4

Track I: Masterpiece: Autonomous Systems

Machines that go where people shouldn't, and come back with answers.

#	Codename	Project	Level	Months
51	SEAWALL	Autonomous survey boat for river and flood-plain mapping	#####	18
52	FLEETMIND	Heterogeneous robot fleet for disaster mapping	#####	20
53	EDGEOS	Secure, updatable OS image for edge-AI fleets	#####	14
54	ROADSAGE	Driver-assistance research stack with a safety monitor	#####	16
55	RAILWATCH	Autonomous rail-track inspection trolley	#####	16
56	AGRIFLEET	Autonomous vineyard robot for mechanical weeding	#####	20
57	PIPEPROBE	In-pipe inspection crawler for sewers	#####	14
58	SKYLANE	Drone traffic management simulator (U-space)	#####	14
59	RECYCLR	Tactile robot arm that sorts recyclables	#####	16
60	MESHLIFE	Disaster mesh network with AI message triage	#####	14

Track J: Masterpiece: Foundation Models and AI Infrastructure

Build the engine, not just the car.

#	Codename	Project	Level	Months
61	TINYFOUNDRY	Train a small language model from scratch, end to end	#####	12
62	AERIALVLM	Vision-language model for disaster aerial imagery	#####	14
63	INFERNO	LLM inference server built from scratch	#####	14
64	MINICOMPILER	Tiny ML compiler for edge accelerators	#####	16
65	RAGOPS	Multi-tenant enterprise RAG platform	#####	14
66	AGENTGRID	Verified multi-agent workflow runtime	#####	14
67	EDGEBENCH	Benchmark and leaderboard for on-device LLMs	#####	12
68	FEDERATE	Federated learning across edge devices with privacy guarantees	#####	14
69	SYNTHSCENE	Synthetic data engine for rare disaster scenes	#####	14
70	HOMECLUSTER	Small GPU cluster with a real scheduler and MLOps	#####	12

Track K: Masterpiece: Trust, Security and Sovereignty

Prove it, don't promise it.

#	Codename	Project	Level	Months
71	SOVEREIGN	EU sovereign AI reference stack, on-prem	#####	18
72	ENCLAVE	Confidential AI inference with remote attestation	#####	14
73	ZKAUDIT	Zero-knowledge proofs for AI compliance claims	#####	16
74	AUTOSOC	Human-gated autonomous security operations centre	#####	16
75	MODELCHAIN	Signed, provenance-tracked ML model supply chain	#####	12
76	DPVAULT	Differential privacy analytics platform	#####	12
77	PROVENANCE	Content provenance and deepfake detection for newsrooms	#####	14
78	PROVENGUARD	Formally verified guardrail policy engine	#####	18
79	OTSENTINEL	Industrial control system anomaly detection lab	#####	14
80	CREDMESH	Verifiable credentials for responders at disaster sites	#####	12

Track L: Masterpiece: Planet, Climate and Infrastructure

Model the world well enough to protect it.

#	Codename	Project	Level	Months
81	FLOODTWIN	Hydrodynamic digital twin of the Ahr valley with an ML surrogate	#####	20
82	GRIDBRAIN	Municipal grid-edge forecasting and battery dispatch	#####	14
83	PLUMEWATCH	Methane plume detection from satellites	#####	14
84	HEATSHIELD	City heat-wave early warning and vulnerability map	#####	12
85	LEAKNET	Water network leak detection with acoustic sensors and graph ML	#####	16
86	DROUGHTSENSE	Drought and soil-moisture forecasting from radar satellites	#####	16
87	BRIDGESHM	Structural health monitoring with vibration sensors and a digital twin	#####	16
88	FIRESPREAD	Wildfire spread surrogate and resource planner	#####	16
89	SURGECAST	Storm-surge forecasting for the North Sea coast	#####	12
90	KMWEATHER	Kilometre-scale weather downscaling with ML	#####	16

Track M: Masterpiece: Civic, Science and Human Performance

Big skills, pointed at real people.

#	Codename	Project	Level	Months
91	ATHLETEOS	Sprint and hurdle performance platform	#####	14
92	COMMANDSIM	Incident command training simulator with AI role players	#####	14
93	VERWALTUNG	Citizen navigator for German administrative procedures	#####	14
94	SKYEDGE	Edge AI for amateur telescope transient detection	#####	14
95	BABEL	Offline speech translation device for responders	#####	14
96	WAYFINDER	Indoor navigation for blind and low-vision users	#####	16
97	REPROBOT	Research agent that reproduces ML papers, with verification	#####	16
98	QUESTLINE	Adaptive, gamified engineering curriculum engine	#####	14
99	HELPERHUB	Coordination platform for spontaneous disaster volunteers	#####	14
100	CITYTWIN	Open digital twin of a mid-size city	#####	24

Track N: Frontier: Superintelligence Alignment and Safety

Nobody builds superintelligence in a garage. Everybody who matters is hiring people who can make it safe.

#	Codename	Project	Level	Months
101	GLASSBOX	Mechanistic interpretability of a small open model with sparse autoencoders	#####	12
102	CONTROLROOM	AI control protocols for untrusted coding agents	#####	12
103	SANDBAG	Detecting sandbagging in capability evaluations	#####	12
104	MODELORGS	Model organisms of emergent misalignment	#####	14
105	WEAKJUDGE	Scalable oversight with debate and weak-to-strong supervision	#####	12
106	EVALSMITH	New safety evaluations contributed to Inspect Evals	#####	12
107	TRUTHPROBE	Lie and deception detection with activation probes	#####	12
108	REWARDGUARD	Catching reward hacking in RL-trained coding agents	#####	14
109	COTWATCH	Chain-of-thought monitorability study	#####	12
110	COMPUTEWATCH	Technical AI governance with hardware-backed compute verification	#####	14

Track O: Frontier: Quantum Computing and Quantum AI

Honest quantum: beat the classical baseline or say so.

#	Codename	Project	Level	Months
111	QDECODER	Neural decoders for quantum error correction	#####	12
112	QMLTRUTH	Honest quantum machine learning benchmark	#####	12
113	QROUTE	Quantum optimisation for disaster logistics, against strong classical solvers	#####	12
114	QMITIGATE	Error mitigation contributions and real-hardware benchmarks	#####	12
115	QCHEM	Quantum chemistry for battery materials, with honest classical comparison	#####	14
116	QCOMPILE	Quantum circuit optimisation passes	#####	12
117	QKDNET	Quantum key distribution network simulator for critical communications	#####	12
118	QSIMGPU	GPU-accelerated quantum circuit simulator	#####	14
119	QAGENT	AI agent that writes and verifies quantum programs	#####	12
120	QREADY	When will quantum matter? Resource estimates for real national problems	#####	12

Track P: Grand Master: Nation-Scale Sovereign AI

Everything in this atlas, assembled for a whole country.

#	Codename	Project	Level	Months
121	NATIONSTACK	Sovereign AI ecosystem for a nation of 100 million people	#####	36

How this repo works

   tracks/*.yaml  --(python tools/render.py)-->  projects/*.md + this index
                                     |
                         CI: tests/ + render --check

Edit a brief in tracks/, run python tools/render.py, commit both. CI fails if the pages and the data drift apart.

Ground rules for every project

• Measured numbers only: every metric has a date, hardware and git commit.
• Public or synthetic data; data card before training.
• Civilian use. Defensive security only, on systems you own or are allowed to test.
• Honest limits section in every README.

Licence

Apache-2.0. Datasets and services named in the briefs keep their own licences.