Candidate Projects — 15 Ideas Scored for Viral × Feasible × Useful

All 15 candidates were generated from 3 lenses (AI-augmented, DX/joy, universal pain) and scored on 4 axes. One winner is picked at the end.

---

Scoring Rubric

Each candidate scored 1–10 on:

Axis	Question
Viral (V)	Would you share this within 5 min of trying it? Is the demo GIF jaw-dropping?
Feasible 5d (F)	Can 1 person build a lovable MVP in 5 all-day sessions in Python?
TAM (T)	How many programmers hit this pain weekly? (10 = every dev, 1 = niche)
Novelty (N)	How underserved is it? (10 = no good solution, 1 = crowded)

Total = V × F × T × N is not literal multiplication — we use weighted composite Score = (V*0.35 + F*0.25 + T*0.25 + N*0.15) scaled to 100. Rankings are what matter.

---

Lens A — AI-Augmented Productivity (5 candidates)

A1 — context / packit — Smart Codebase Packer for LLMs

Field	Detail
One-liner	packit . --ask "where is auth?" → copies only relevant files to clipboard, token-counted
Problem	Pasting code into Claude/Cursor is manual, hits token limits, includes junk
Solution	AST + embeddings (local, via sentence-transformers) to find relevant files; TUI file picker with live token counter; .gitignore-aware; copy to clipboard
Demo hook	Split screen: manual copy-paste hell vs packit --ask in 2 sec
Tech	Python, sentence-transformers (all-MiniLM), rich, pyperclip, tiktoken
Differentiator vs gitingest/repomix	Question-aware retrieval + interactive TUI + live token budget
Monetization	Pro: team shared contexts, cloud history
Scores	V 8 · F 9 · T 10 · N 5 → Composite 81
Verdict	Strong, but crowded — gitingest/repomix already viral, need 10× to displace

A2 — diffsense — Semantic PR Explainer

Field	Detail
One-liner	diffsense in a PR branch → beautiful terminal summary of what actually changed (not just lines)
Problem	GitHub diff is syntactic; reviewers miss semantic changes ("this refactors auth but breaks rate limiting")
Solution	Parse diff + AST diff + optional LLM to generate "what changed, why it matters, risks" summary
Demo hook	Ugly GitHub diff vs beautiful diffsense output with risk highlights
Tech	Python, unidiff, tree-sitter, rich, optional OpenAI API
Scores	V 7 · F 7 · T 8 · N 6 → 70
Verdict	Useful but PR-specific — narrower TAM than codebase-wide

A3 — aigen — AI Test Generator That Doesn't Suck (Local)

Field	Detail
One-liner	aigen src/auth.py → generates tests/test_auth.py with high-coverage pytest tests
Problem	Writing tests is hated; AI test generators exist but produce flaky/low-coverage tests
Solution	AST-aware prompt engineering + mutation testing to self-correct; generates only tests that pass locally before showing
Demo hook	ls tests/ (empty) → aigen → pytest (all green, 85% coverage)
Tech	Python, ast, coverage.py, mutmut ideas, LLM API
Scores	V 7 · F 5 · T 9 · N 5 → 65
Verdict	Hard to get quality right in 5 days — flaky demo kills viral

A4 — promptlint — Linter for LLM Prompts in Code

Field	Detail
One-liner	promptlint . finds brittle prompts in your codebase and suggests robust rewrites
Problem	Prompts embedded in code are untested, unversioned, and break silently
Solution	Detect prompt strings (heuristic + AST), lint for anti-patterns, suggest improvements via LLM
Demo hook	"Found 12 prompts — 3 are brittle" with inline suggestions
Tech	Python, ast, rich, LLM API
Scores	V 5 · F 7 · T 5 · N 8 → 60
Verdict	Too niche — only teams with heavy prompt usage feel it

A5 — hallucatch — Catches AI-Hallucinated APIs

Field	Detail
One-liner	hallucatch app.py flags calls to non-existent APIs/packages that AI likely hallucinated
Problem	AI generates import nonexistent_lib or client.fake_method() that fails at runtime
Solution	Static analysis + PyPI index check + type stub awareness to flag hallucinated imports/calls
Demo hook	AI-generated file with 3 hallucinated calls → hallucatch highlights them in red
Tech	Python, ast, importlib.metadata, PyPI API, rich
Scores	V 6 · F 8 · T 7 · N 7 → 69
Verdict	Clever, but hallucination rate is dropping as models improve — shrinking TAM

---

Lens B — Pure DX & Joy (5 candidates)

B1 — wtf / oops / autopsy — Beautiful Traceback That Explains Itself

Field	Detail
One-liner	wtf python app.py — any crash becomes a beautiful, plain-English explanation + fix suggestion
Problem	Python tracebacks are still cryptic; juniors stare at them, seniors waste time
Solution	Rich traceback + heuristic explainer (offline, no API key) + optional LLM deep-explain; wtf --fix applies patch
Demo hook	Side-by-side: CPython ugly traceback vs wtf beautiful explained output — night and day
Tech	Python, rich, traceback, ast, optional LLM
Scores	V 9 · F 9 · T 9 · N 6 → 84
Verdict	Strong runner-up. Massive TAM, amazing demo, 5-day feasible. Weakness: rich traceback already exists, need to 10× it

B2 — lumen / spy — Inline Execution Visualizer

Field	Detail
One-liner	lumen app.py — see variable values inline as code runs, no debugger, no print statements
Problem	print() debugging is manual; pdb is hostile; snoop/icecream not visual enough
Solution	sys.settrace hook that renders variable flow inline in terminal with colors/arrows; lumen --watch x,y
Demo hook	Mesmerizing GIF: code on left, values flowing on right, loop variables animating
Tech	Python, sys.settrace, ast, rich/textual
Scores	V 10 · F 6 · T 7 · N 7 → 77
Verdict	Most visually viral, but sys.settrace edge cases (async, C extensions) make 5-day polish hard

B3 — peek / atlas / cartographer — Understand Any Codebase in 5 Seconds ⭐

Field	Detail
One-liner	peek . → beautiful interactive architecture map of any repo in 5 seconds
Problem	Cloning a repo → "what does this do? where to start?" → 30 min of find/grep/README
Solution	AST import graph + file ranking + heuristic summarizer + Textual TUI; optional LLM summary; HTML export
Demo hook	git clone k8s → peek . → stunning map: modules, entry points, "start here"
Tech	Python, ast, pathlib, rich, textual, networkx (optional), LLM optional
Scores	V 10 · F 8 · T 10 · N 7 → 90
Verdict	WINNER. Highest composite. Universal, visual, 5-day feasible, underserved.

B4 — ship / commitcraft — One-Command Ship: Commit + Push + PR

Field	Detail
One-liner	ship — analyzes git diff → writes perfect commit message → pushes → creates PR with description
Problem	Writing commit messages and PR descriptions is daily toil, done poorly
Solution	git diff → heuristic + optional LLM → conventional-commit message + PR body; beautiful preview + ship --yolo
Demo hook	Messy diff → ship → beautiful commit + PR in 2 sec, side-by-side
Tech	Python, gitpython/subprocess, rich, LLM API
Scores	V 8 · F 9 · T 10 · N 4 → 79
Verdict	Huge TAM, easy to build, but crowded (aicommits, opencommit, commitgpt exist — low novelty)

B5 — dive / loglens — Magical Log Tailer

Field	Detail
One-liner	dive app.log — auto-detects log format, colorizes, filters, and explains errors inline
Problem	tail -f is dumb; logs are unstructured; finding errors is manual grep
Solution	Heuristic log parser + pattern detection + error highlighting + dive --errors-only + time histogram
Demo hook	Wall of grey logs vs dive — errors pop in red, histogram at bottom
Tech	Python, rich, textual, regex heuristics
Scores	V 7 · F 8 · T 7 · N 6 → 71
Verdict	Useful for backend/devops niche, less universal than peek

---

Lens C — Universal Workflow Pain (5 candidates)

C1 — envdoctor — Python Env Doctor & Fixer

Field	Detail
One-liner	envdoctor --fix — diagnoses and fixes venv/pip/python version issues automatically
Problem	"Works on my machine" + cryptic pip errors + venv confusion wastes hours
Solution	Scan pyproject.toml, requirements*.txt, venv, python --version, pip check → diagnose + auto-fix
Demo hook	Broken env with 5 errors → envdoctor → "Fixed 4/5, here's the last one"
Tech	Python, packaging, importlib.metadata, subprocess, rich
Scores	V 6 · F 8 · T 8 · N 6 → 70
Verdict	High pain but demo is not visually viral — hard to GIF

C2 — findit — Semantic Code Search ("find function by what it does")

Field	Detail
One-liner	findit "where is auth token validated" → finds relevant code by intent, not keyword
Problem	grep -r is keyword-only; you need to know the function name to find it
Solution	Local embeddings of code chunks (docstrings + names + bodies) via sentence-transformers; findit "query" → ranked results
Demo hook	grep "auth" (10 noisy hits) vs findit "auth token validated" (1 perfect hit)
Tech	Python, sentence-transformers, ast, rich
Scores	V 7 · F 7 · T 9 · N 7 → 75
Verdict	Strong, but embedding setup adds friction (model download); could be a peek feature instead

C3 — mockit — Instant API Mock Server

Field	Detail
One-liner	mockit --record https://api.example.com captures traffic → mockit --serve serves mocks locally
Problem	Frontend/dev without backend, testing without staging, flaky external APIs
Solution	HTTP proxy recorder + FastAPI mock server from OpenAPI or captured traffic
Demo hook	Real API call → mockit records → offline mock serves instantly
Tech	Python, FastAPI, httpx, mitmproxy concepts
Scores	V 7 · F 6 · T 7 · N 5 → 64
Verdict	Crowded (msw, wiremock, prism) — hard to differentiate in 5 days

C4 — depcruise / depgraph — Beautiful Dependency Visualizer

Field	Detail
One-liner	depcruise . → interactive dependency graph of your Python project (imports + pip deps)
Problem	pipdeptree is text-only; import cycles and bloat are invisible
Solution	AST imports + importlib.metadata → graph with cycle detection, bloat highlights, depcruise --why package
Demo hook	pipdeptree (wall of text) vs depcruise (beautiful interactive graph, cycles in red)
Tech	Python, ast, networkx, rich/textual, graphviz optional
Scores	V 7 · F 8 · T 7 · N 6 → 71
Verdict	Could be a killer feature of peek, but standalone is narrower

C5 — gitlens-lite — Git History Visualizer for Humans

Field	Detail
One-liner	gl file.py — see who changed what, when, why — beautifully, in terminal
Problem	git log/blame are hostile; understanding file history is painful
Solution	git log --follow + blame → rich timeline, author viz, "why was this line added?"
Demo hook	git log --oneline (wall of hashes) vs gl (colored timeline with messages)
Tech	Python, gitpython/subprocess, rich
Scores	V 6 · F 8 · T 7 · N 5 → 65
Verdict	lazygit/gitui already cover this well enough

---

Composite Ranking — Top 15

Rank	ID	Name	One-liner	V	F	T	N	Composite	Tier
1	B3	peek	Understand any codebase in 5 sec	10	8	10	7	90	🏆 Winner
2	B1	wtf	Beautiful traceback that explains itself	9	9	9	6	84	Runner-up
3	A1	packit	Smart codebase packer for LLMs	8	9	10	5	81	Runner-up
4	B4	ship	One-command commit+push+PR	8	9	10	4	79	Strong
5	B2	lumen	Inline execution visualizer	10	6	7	7	77	Visual king, hard
6	C2	findit	Semantic code search	7	7	9	7	75	Feature of peek
7	A2	diffsense	Semantic PR explainer	7	7	8	6	70	Narrow
7	C1	envdoctor	Env doctor & fixer	6	8	8	6	70	Not viral
9	B5	dive	Magical log tailer	7	8	7	6	71	Niche
9	C4	depcruise	Dependency visualizer	7	8	7	6	71	Feature of peek
11	A5	hallucatch	Catches hallucinated APIs	6	8	7	7	69	Shrinking
12	C3	mockit	Instant mock server	7	6	7	5	64	Crowded
12	A3	aigen	AI test generator	7	5	9	5	65	Flaky demo
14	C5	gitlens-lite	Git history viz	6	8	7	5	65	Covered
15	A4	promptlint	Prompt linter	5	7	5	8	60	Niche

Why peek won — the 4-way test

Test	peek	Next best (wtf)	packit
Is demo GIF jaw-dropping?	✅ Architecture map animating	✅ Traceback before/after	⚠️ TUI file picker — good not great
Does every dev feel this weekly?	✅ Every repo clone/onboarding	✅ Every crash	✅ Every AI coding session
Can it work with zero config?	✅ peek . — no API key needed	✅ Offline heuristics	⚠️ Needs embeddings download
Is competition weak?	✅ No beautiful local codebase map	⚠️ rich traceback exists	❌ gitingest/repomix already viral
Shareability	✅ Screenshot of any repo's map	✅ Side-by-side traceback	⚠️ Clipboard action — invisible

peek is the only candidate that scores ✅ on all five.

---

What to do with runners-up

Don't discard — absorb them as peek features (post-MVP):

Runner-up	How it becomes a peek feature
wtf traceback	peek --explain traceback.txt
packit smart packer	peek --pack --ask "where is auth?"
findit semantic search	peek --find "auth validation"
depcruise deps	peek --deps
ship commit	peek already maps repo → helps write better commits

This makes peek a platform for codebase understanding, not a one-trick tool — compounding viral surface area.

---

Next → 04_project_details__peek.md — the full spec for the winner.

---

Author: Hariom Lohar -- hariomlohar.new@gmail.com -- https://hariomlohardev.github.io/ -- 2026-08-10