The AI memory layer for DSA

failed.Here's exactlywhy.

memoised turns the AI you already pay for — Claude, ChatGPT, Gemini — into a DSA coach. It reads every submission, tells you what broke, and makes sure you never fail that way twice.

Add to Claude→See how it works
~/.config/claude/mcp.jsonno install · one line
"memoised": {
  "type": "streamable-http",
  "url": "https://dsa.memoised.dev/mcp"
}
claude
memoised connected
give me a graph problem I haven't solved — medium
→ get_problem(category="graphs", difficulty="medium")
Course Schedule IIMEDIUM
Return an order to finish all courses given prerequisites. Written to ~/dsa/course_schedule.py
done — submit
→ submit_solution() · ran 12 tests
ACCEPTED12/12 · 38ms
Clean topological sort. Re-queued in 6 days to lock it in.
nice. give me a harder one
→ get_problem(category="graphs", difficulty="hard")
Alien DictionaryHARD
Infer letter order from a sorted list of alien words. Written to ~/dsa/alien_dict.py
WORKS INClaude DesktopClaude CodeCursorVS CodeNeovim

THE MEMORY LAYER FOR YOUR DSA

Your AI forgets every chat.
memoised remembers.

Your assistant starts every conversation from zero. memoised is the memory layer underneath it — every submission tagged with its mistake and code habits, so your weak spots compound into a profile that only gets sharper the more you code.

7 + 10
mistake & habit categories
SM-2
spaced-repetition engine
0 tabs
it's all in your editor
your memory
writing…
Two SumACCEPTED
Subarray Sum KOFF_BY_ONE
magic_numbersHABIT ↑
Course Schedule IIACCEPTED
Koko BananasTLE
Word LadderRE-QUEUED
Binary SearchOFF_BY_ONE
single_letter_varsIMPROVING
Find PeakACCEPTED
Two SumACCEPTED
Subarray Sum KOFF_BY_ONE
magic_numbersHABIT ↑
Course Schedule IIACCEPTED
Koko BananasTLE
Word LadderRE-QUEUED
Binary SearchOFF_BY_ONE
single_letter_varsIMPROVING
Find PeakACCEPTED
nothing here ever resets →

THE GAP

Same wrong answer. Two different worlds.

// the old way
✗ Wrong Answer
Runtime: — · 82 / 211 cases
That's it. That's the entire feedback.
m
memoised spotted a repeat
OFF_BY_ONE
6 of your last 20+ subs

Seems like across your last 20+ submissions, you keep making this same off-by-one. Here, your binary search uses lo < hi but updates hi = mid — so the final index never gets checked. Same slip you made on Find Peak and Koko Bananas.

✓Avoid it next time: when the answer can be the last element, your loop has to be able to reach it — reach for lo <= hi.

PATTERN RECOGNITION

You don't fail randomly.
You fail the same way.

memoised catches the mistake you keep repeating, names it in plain English, and drills you through the problems that train it out — until it stops happening.

memoised · pattern engine
updated every submission

01 · WE SEE IT

your last 20 submissions →

6 of your last 20 fails trace back to the exact same bug.

→

02 · WE NAME IT

Not just “Wrong Answer.” Every failure gets a real name:

OFF_BY_ONEEDGE_CASE_MISSEDRECURSION_BASE_CASE_MISSEDBAD_VARIABLE_NAMEWRONG_COMPLEXITYMISSING_NULL_CHECK

OFF_BY_ONE — yours: the loop bound stops one step short, every time.

→

03 · IT FADES

off-by-one, per attempt →

memoised drills it until it stops happening.

WHAT MAKES IT AI-NATIVE

Bring the AI you already pay for.

memoised isn't another chatbot with another subscription. It's an MCP tool that plugs into the assistant you already own — so Claude, ChatGPT, or Gemini becomes your DSA tutor. You bring the brain. memoised brings the problems, the sandbox, and the memory.

YOU ALREADY PAY FOR

ClaudePro
ChatGPTPlus
GeminiAdvanced

Your tutor — explains problems, ladders hints, debugs with you.

+

YOU ADD (ONE LINE)

memoised(dsa)

remote MCP · no install · no new AI bill

The DSA brain — problems, sandbox, mistake tags, memory.

=

YOU GET

A coach that lives in your chat.

memoised connected

Practice that understands how you fail — right where you already work.

Your assistant handles the talking
→Turns "give me a graph problem I keep failing" into a real fetch
→Explains, in plain language, exactly why your code broke
→Ladders hints from a nudge to the full approach — when you ask
→Walks the optimal solution and debugs with you, in context

// on your existing subscription — memoised never resells AI

memoised handles the thinking
✓A curated, embeddable problem bank with hidden test cases
✓Runs your code in a real sandbox and returns the true verdict
✓Tags every failure into a mistake taxonomy & habit profile
✓Schedules revision with SM-2 and tracks your patterns over time

// the data & memory layer your chatbot can't have alone

THE WORKFLOW

Practice where you already code.

No browser tab. No copy-paste. Three tool calls and you're in a loop.

01

Ask Claude

“Give me a graph problem I keep failing.” Claude pulls one matched to your weak topics.

get_problem
02

Solve in your editor

Starter code lands in your repo. Write it in VS Code, Neovim, Cursor — your real setup, your keybinds.

your filesystem
03

Submit & understand

Claude runs it in a sandbox, then explains the verdict, the mistake, and when to revisit it.

submit_solution

WHAT YOU GET

A coach that actually remembers you.

Mistake taxonomy

Every failure is tagged into one of 7 categories — off-by-one, wrong complexity, edge case — not just “wrong”.

tagged on every failure

Code habit profile

Tracks 10 recurring habits — magic numbers, god functions, reinventing stdlib — across every language you use.

reviewed as you go

Spaced repetition

An SM-2 queue resurfaces problems right before you'd forget them. Fail it → it comes back tomorrow.

get_revision_queue

Pattern analytics

Weak-topic scores, mistake frequencies, and a 30-day trend of exactly where you bleed points.

get_analytics

Progressive hints

Stuck? Four escalating hint levels — from a nudge to the full approach — so you're never just staring.

four escalating levels

Semantic search

“Problems like the one I just failed.” memoised finds conceptually similar ones you haven't solved yet.

meaning, not keywords

Practice DSA the way 2026 interviews actually work.

With AI in your workflow — understanding your patterns, getting smarter every submission.