Remix

Make your own version of Student Query Understanding

Edit the prompt below and generate a customized version in seconds.

Original by BHUVANESWARI S CSE-2023

Describe your app

============================================================
NLP / NATURAL LANGUAGE UNDERSTANDING
============================================================

Add an NLP/NLU layer using:

RAG/query_processor.py

The QueryProcessor must understand the meaning of student questions
before retrieval instead of depending on exact keywords or a large
hardcoded ALIASES dictionary.

Workflow:

User Question
    ↓
NLP Processing
    ↓
Normalization + spelling/grammar handling
    ↓
Intent + Entity + Department + Year detection
    ↓
Question Type Detection
    ↓
Canonical Query
    ↓
Embedding
    ↓
Hybrid Retrieval

NLP MUST HANDLE:

- Different sentence structures
- Synonyms and paraphrases
- Short/informal questions
- Grammar mistakes
- Minor spelling mistakes
- Abbreviations
- Department names
- Intent
- Years such as 2024/2025/2026
- "last year", "latest", "previous year", etc.
- Follow-up questions

EXAMPLE:

"Who is the HOD of CSE?"
"Who heads CSE?"
"who is leading computer science?"
"cse hod?"

Should produce approximately:

intent = hod
department = Computer Science and Engineering
question_type = person_lookup
canonical_query = Head of Department Computer Science and Engineering

ADMISSION EXAMPLE:

"how can I join NEC?"

→ intent = admission
→ question_type = procedure
→ canonical_query = NEC admission process

PLACEMENT EXAMPLE:

"how was placement last yr?"

→ intent = placement_statistics
→ temporal_reference = previous_year
→ canonical_query = NEC placement statistics previous year

DEPARTMENT NORMALIZATION:

CSE → Computer Science and Engineering
ECE → Electronics and Communication Engineering
EEE → Electrical and Electronics Engineering
IT → Information Technology
AI/DS/AIDS → Artificial Intelligence and Data Science

Use NLP/entity normalization rather than hundreds of aliases.

INTENT TYPES:

admission
eligibility
courses
department
hod
faculty
placement
placement_statistics
recruitment
package
hostel
transport
scholarship
facilities
library
sports
examination
results
training
clubs
contact
administration
general
unknown

QUESTION TYPES:

person_lookup
procedure
list
statistic
date
location
contact
explanation
comparison
yes_no
eligibility
fact_lookup

QUERY PROCESSOR OUTPUT:

{
    "original_query": "...",
    "normalized_query": "...",
    "canonical_query": "...",
    "intent": "...",
    "question_type": "...",
    "department": "...",
    "entities": [],
    "year": null,
    "temporal_reference": null,
    "keywords": [],
    "confidence": 0.0
}

IMPORTANT:

NLP must NOT replace semantic retrieval.

The final system must be:

User Question
      ↓
NLP/NLU
      ↓
Intent + Entity + Department + Year
      ↓
Canonical Query
      ↓
Embedding
      ↓
FAQ + Pinecone Retrieval
      ↓
Hybrid Re-ranking
      ↓
Top Relevant Chunks
      ↓
Ollama
      ↓
Grounded NEC Answer

Do NOT use a huge ALIASES dictionary.

If NLP confidence is low, do not force an incorrect interpretation.
Fall back to semantic + keyword retrieval.

The goal is:

Question
→ Understand meaning
→ Find correct NEC evidence
→ Generate accurate grounded answer.

Free to use. No credit card required.