Retail journey

The “add” tab — six modes, one memory.

Every Mukodi user · across every life stage and surface

Whether a user types, speaks, snaps, files, forwards, or asks — Mukodi reads the input, classifies the intent, and writes one canonical memory. Six modes, one ledger.

Marketing claimingestion-tiers
Try it in the app Talk to us
The story · 6 scenes

Beginning to end.

Every Mukodi user · across every life stage and surface
·Whether a user types, speaks, snaps, files, forwards, or asks — Mukodi reads the input, classifies the intent, and writes one canonical memory. Six modes, one ledger.
  1. Scene 01 · Type a transaction in plain language
    “Paid Anita ₹1500 for groceries.” The categoriser tags it L1 / L2, infers the merchant, computes tax columns, and writes one row to the ledger.
  2. Scene 02 · Speak in any of eight languages
    The same input in Tamil or Hindi lands at the same row — the STT pass routes through the intent classifier and the categoriser the same way.
  3. Scene 03 · Snap a bill
    An image of a printed bill is OCR'd by the VLM; line items are matched to merchants and split into tax columns.
  4. Scene 04 · Add a file — CSV, parquet, OFX, MT940, statement PDF, ZIP
    Any of 36 supported formats is detected, mapped, and folded into the ledger. Bank-statement PDFs are table-extracted directly.
  5. Scene 05 · Forward an SMS
    A bank SMS is parsed for the amount, the merchant, and the leg type. The same row lands in the ledger with the SMS body kept on file for replay.
  6. Scene 06 · Ask a question that adds nothing
    “How much did I spend on groceries last month?” reads the same ledger and replies in the same language the user asked. No new row, one memory.
Helpers on call
The stack · helpers + models

Who's actually doing the work.

Features used
Transactions & insightsConversational AI advisor8-language voice (STT + TTS)36-format ingest (Tier 1-5)Documents & KYCSMS forward parserYour memory, your rules
Honest limits

What this flow doesn't do.

  • OCR confidence on snapped bills depends on lighting and resolution; the app surfaces a confidence band and lets the user edit before save.
  • SMS ingestion requires the user to forward the SMS or grant the read-SMS permission; Mukodi never silently reads SMS without explicit consent.
  • The 36 ingestion formats include audio and video for knowledge / labels uploads; transaction uploads use the tabular and bank-statement subset.
← Back to all use cases