Sound-to-Spelling Graph
"Which letters make the same sound?"
The child links what they hear to what they see. fan, phone, laugh, off all carry one sound — spelled four ways. Nodes are sounds; links are their spelling disguises.
Language, taught by discovery — grade by grade
Vaaani leads each child to discover the sounds, word families and meanings of English for themselves — pitched to their grade, bridged from their home language, and grown one student at a time. Never handed the answer; always led to it.
Most AI hands you the answer. Vaaani asks you back — anchored in the very text you're trying to understand.
Every word the child meets becomes a node.
Whenever two words share a hidden piece or a meaning, a link connects them.
Families of related words cluster into worlds — automatically.
The child does not just memorise words. They decode them.
One graph · three stages · every grade
Not three separate tools — one living web that unlocks a new layer as the student is ready, and adapts to where each child actually is. The grown-up names (root, phoneme, etymology) stay hidden until the child has discovered the idea for themselves.
"Which letters make the same sound?"
The child links what they hear to what they see. fan, phone, laugh, off all carry one sound — spelled four ways. Nodes are sounds; links are their spelling disguises.
"What little piece is hiding inside them all?"
The child discovers the shared piece that powers a family. triangle, tricycle, tripod — the piece tri means three. Nodes are words; links are the piece they share.
"How do the families connect?"
Families join families. Pieces combine into new words, and the grown-up names finally appear. The whole web of meaning — and only now, the science behind it.
Student-specific, not class-average: Vaaani remembers each child — which pieces they’ve uncovered, where they paused, which home language (Hindi, Bangla…) they bridge from — and grows that child’s graph at their own pace.
The same discovery engine, tuned to each grade band — sounds first, then word families, then meaning and sentences. Scroll → to walk through them.
How it works
Vaaani turns whatever you upload into a navigable map of ideas — then teaches across that map.
PDFs, notes, essays, problem sets, DOCX, plain text. Vaaani chunks each document, embeds it, and indexes everything locally.
Every word becomes a node. Every time two words connect through etymology or meaning becomes a link. Roots, derivatives, and linguistic domains cluster automatically.
Get clear answers grounded in your sources. Or flip Socratic mode on and Vaaani teaches by asking — never spoiling the punchline.
Grade-aware pedagogy
One discovery engine, four strands — pitched to the grade in front of it. A Grade 2 child meets sounds and word families; the science and the grown-up names wait until they're ready.
Asks how sounds are produced, what the place and manner of articulation are, and why a sound behaves differently across languages. Will not tell you the IPA symbol until you describe the articulators involved.
Leads the child to spot the little piece hiding inside a family of words, then work out what it means — by discovery, never by being told. The grown-up names (root, prefix, suffix) are revealed only once the child has earned them.
Points to sentence structure and grammatical rules. Asks why one construction works and another does not. Always anchors explanation in syntactic theory — not just "it sounds wrong."
Connects word families into a web of meaning and bridges each piece to the child's home language (Hindi, Bangla, and more). Older students unlock the science — how one ancient piece seeded words across languages they already speak.
What's inside
Everything runs on your machine. Embeddings, the vector index, the graph — local. Only the language model call leaves the box.
4-bit quantised vectors. A million-chunk corpus fits in roughly 4 GB of RAM, with faster search than uncompressed indexes.
Every chunk contributes entities and relations. Global queries map-reduce across community summaries — not against a single prompt window.
In non-Socratic mode every sentence is checked against retrieved context. Ungrounded claims are detected before they reach you.
Persistent facts and recent queries inject themselves into prompts when relevant — without bloating context for unrelated questions.
If your corpus doesn't contain the answer, Vaaani says so. It will not invent material to look helpful.
Every wrong answer is classified — spelling, careless slip, partial-recall, conceptual gap — so the student sees the shape of their mistakes, not just a score. Open the X-Ray
Timed mock papers with confidence-rating after every answer. Vaaani grades against your corpus, surfaces where your nerves cost marks. Start a session
Discover the little piece hiding inside a family of words, work out what it means, then decode words you've never met instead of cramming them. Colour-coded so you see how each word is built. Free for everyone. Grow a word family
Tap any phonetic symbol to hear it, see how the mouth makes it, and learn which sounds Indian-English speakers most often swap. Free for everyone — no sign-in. Explore the sounds
Active practice for the sound system: change one feature of a sound — voice, place, manner — and hear it become another, exactly as the Grade-2 lessons teach. Each connection you make fills in the IPA chart you're learning. Open the Sound Lab
Launch Vaaani, upload your first PDF, and ask it anything about what you just uploaded.