Symbiote build18 books · 4,551 pages

YOUR TEXTBOOK

JUST WOKE UP

SSC-এর জন্য একটি AI শিক্ষক, যা শুধু NCTB নবম-দশম শ্রেণির বই থেকে উত্তর দেয় — বই ও পৃষ্ঠা ধরে উদ্ধৃত করে, আর বইয়ে না থাকলে সরাসরি বলে দেয়।

See the evidence

No sign-up wall · Bangla, English & Banglish · Free to use

◈

Live answer

পদার্থের গতিশক্তি কী? সূত্রসহ বুঝাও।

গতিশক্তি হলো কোনো বস্তুর গতির কারণে অর্জিত শক্তি। ভর m ও বেগ v-এর জন্য —

Ek = ½ m v²

▤ পদার্থবিজ্ঞান p.104✓ grounded

every claim carries its page

◆ 18 NCTB books ingested◆ 4,551 pages◆ hybrid dense + BM25 retrieval◆ reciprocal rank fusion◆ LLM listwise rerank◆ every claim cited to book & page◆ abstains when the books are silent◆ বাংলা · English · Banglish◆ 18 NCTB books ingested◆ 4,551 pages◆ hybrid dense + BM25 retrieval◆ reciprocal rank fusion◆ LLM listwise rerank◆ every claim cited to book & page◆ abstains when the books are silent◆ বাংলা · English · Banglish

Measured, not claimed

The numbers are on the wall

Retrieval is evaluated against a fixed set of 40 questions with known source pages. Hybrid retrieval with reciprocal rank fusion never loses to dense-only and wins clearly at k≥5 — but on 40 questions the difference does not reach statistical significance, and the numbers below say so rather than rounding the caveat away.

Hybrid + RRF vs dense-only retrieval

Gold set, n = 40. Higher is better. Source: backend/data/eval/retrieval_report_v3.json

  • Dense only
  • Hybrid + RRF

+25%

Recall@5 lift from hybrid retrieval

Hybrid wins every disagreement at k=5 (5–0), but with 40 questions that is p = 0.0625 — directional, not yet significant.

70%

Answer relevance

Judged on 12 answers. Faithfulness to the retrieved context: 50%.

5/6

Correct refusals

On out-of-syllabus probes it declines instead of inventing an answer.

Small-sample figures — 40 retrieval questions, 12 scored answers, on a pinned question set with the answer cache disabled. McNemar's exact test puts the hybrid-vs-dense gap at p = 0.0625 (k=5): consistent, not yet significant. Reported as-is rather than rounded up. Full benchmark console →

How an answer gets made

Six panels, one answer

  1. 01

    Question

    Bangla, English or Banglish. Intent, subject and grade are parsed off the raw text.

  2. 02

    Expand

    HyDE writes a hypothetical answer; its embedding widens the net for vocabulary the student didn't use.

  3. 03

    Retrieve ×2

    Dense vectors and BM25 run in parallel over parent-child chunks of the corpus.

  4. 04

    Fuse

    Reciprocal rank fusion merges both rankings — no hand-tuned weight to overfit.

  5. 05

    Rerank

    An LLM listwise pass reorders the survivors by actual answer-bearing content.

  6. 06

    Answer

    Generation is constrained to the retrieved context and stamped with [book, page].

The corpus

Every book. Every page.

Open the library

Straight answers

Questions about the questions

এটা কি সত্যিই NCTB বই থেকে উত্তর দেয়?

হ্যাঁ। প্রতিটি উত্তর ১৮টি NCTB বই থেকে retrieve করা হয় এবং বই ও পৃষ্ঠাসহ উদ্ধৃত করা হয়। বইয়ে না থাকলে এটি সেটা স্বীকার করে — বানিয়ে বলে না।

How do I know it isn't making things up?

Two ways. Every claim carries a citation you can open in the library. And the system is measured on an abstention probe set — questions deliberately outside the syllabus — where it correctly refuses 5 of 6 times. The full evaluation is public on the ops console.

কোন ভাষায় প্রশ্ন করতে পারি?

Bangla, English বা Banglish — যেভাবে খুশি। এটি একই ভাষা ও লিপিতে উত্তর দেয়, বাংলার জন্য আলাদা টাইপোগ্রাফিসহ।

Is it free?

Yes. Sign in with Google and start. The whole stack is designed to run on free-tier infrastructure so cost never becomes the reason a student can't use it.

You and the whole syllabus. Together.

WE ARE

SHIKKHA AI