Meeting Notes AI in India | unoblox

A meeting transcript is long, unstructured, and mostly filler — turning it into a short summary with clear action items is a well-suited job for a language model, provided you feed it a decent transcript and ask for a specific output shape rather than "summarise this." unoblox handles the summarisation step over one rupee-billed, OpenAI-compatible endpoint, so it slots behind whatever transcription tool you're already using.

The meeting-notes problem

Recording a meeting solves capture, not comprehension — nobody re-reads a forty-minute transcript to find the three things they agreed to do. Manual note-taking, meanwhile, misses details while the note-taker is busy participating. Automated summarisation sits between the two: full capture from the transcript, distilled output for actually reading afterward.

A summarisation pipeline

1. Transcribe the meeting audio using whatever speech-to-text tool you already use — unoblox's chat models take text in, not raw audio.
2. Chunk long transcripts if they exceed a comfortable prompt size, summarising in sections if needed.
3. Summarise and extract with a clear system prompt: ask explicitly for a short summary, a list of decisions, and a list of action items with owners where mentioned.
4. Distribute the structured output to attendees, or push action items into your task tracker.

curl https://api.unoblox.ai/v1/chat/completions \
  -H "Authorization: Bearer ub-gw-your-key-here" \
  -H "Content-Type: application/json" \
  -d '{
    "model":"deepseek-ai/deepseek-v4.1-flash",
    "messages":[{"role":"system","content":"Summarise the transcript in three bullet points, then list action items with an owner if named."},{"role":"user","content":"<meeting transcript here>"}]
  }'

Choosing a model for transcript length and volume

• Frequent, routine standups: DeepSeek V4 Flash (₹9.07 / ₹18.14 per 1M tokens) keeps daily summarisation cheap.
• Longer or more important meetings: DeepSeek V4.1 Flash (₹20.16 / ₹60.48) or Qwen3 235B-A22B (₹9.07 / ₹55.44) for better handling of nuance across a long discussion.
• Board or client-facing meetings: Claude Sonnet (₹201.6 / ₹1008) or GPT-5 (₹126 / ₹1008) where summary quality matters more than per-call cost.
• Anything else in the catalog, including GPT-4.1 mini and Gemma 4 models, works the same way — check /models for current ₹ pricing.

Privacy considerations for internal meetings

Meeting transcripts often contain sensitive business discussion, so treat model choice as a privacy decision too. unoblox's own small hosted models, such as the free Qwen3 1.7B, process entirely on Indian infrastructure. Third-party models like GPT or Claude process on that provider's own infrastructure — the India advantage there is rupee billing, one GST invoice, and no international card, not data residency — so weigh that against how sensitive a given meeting actually is before choosing which model summarises it.

Frequently asked questions

Does unoblox transcribe audio, or only summarise text? Only text summarisation — you'll need a separate speech-to-text step before sending the transcript to a chat model.

Can it identify who said what? Only if your transcript already labels speakers; the model works with whatever structure is in the text you provide, it doesn't do speaker identification itself.

How long can a transcript be? This depends on the context limits of the model you choose — check the specific model's page under /models, and chunk very long transcripts into sections if needed.

Is this suitable for legally sensitive meetings? Treat it the same as any other document-handling decision — use unoblox's own hosted models for the highest-sensitivity content given the India-infrastructure point above, and apply your normal confidentiality judgement regardless of which model you use.

Can action items be pushed automatically into a task tool? The model can output action items in a structured format (ask for JSON mode); wiring that into your task tracker is a small integration step in your own application.

Does formatting the transcript better improve summary quality? Generally yes — cleaner speaker labels and less filler text give the model less to wade through, which tends to produce a tighter summary.

Get started in rupees → https://unoblox.ai/sign-in