Python Q&A Chatbot Agent
Budget: ₹1,500 – ₹12,500 INR
I’m building a Python-based chatbot that can listen to a user’s question in plain English and return an accurate answer by looking the information up in an existing database. In short, the agent must accept free-form queries, interpret their intent with natural-language processing, fetch the right record through a database lookup, and reply in a clear, human-readable sentence.
Here is what I need from you:
• A clean Python 3 solution. You’re welcome to lean on libraries such as spaCy, Haystack, LangChain, or a lightweight custom approach—whatever achieves solid intent recognition and ranking accuracy.
• Direct integration with my relational database (PostgreSQL at the moment, though I can switch to MySQL if that simplifies your stack).
• A minimal CLI or REST endpoint so I can ask questions locally and see the answers returned. I’ll extend the front end later.
Acceptance criteria
1. All code runs locally with a requirements.txt and concise setup notes.
2. Rephrased but semantically similar questions still locate the correct row(s) in the database.
3. Average response time under two seconds on a table of roughly 50 000 rows.
Once this baseline is working, we can discuss optional refinements such as confidence scoring, answer summarisation, or a caching layer, but the first milestone is a reliable question-answer flow backed by the database. Let me know which NLP tools you prefer, any pretrained embeddings you’ll need, and a rough delivery timeline.
Here is what I need from you:
• A clean Python 3 solution. You’re welcome to lean on libraries such as spaCy, Haystack, LangChain, or a lightweight custom approach—whatever achieves solid intent recognition and ranking accuracy.
• Direct integration with my relational database (PostgreSQL at the moment, though I can switch to MySQL if that simplifies your stack).
• A minimal CLI or REST endpoint so I can ask questions locally and see the answers returned. I’ll extend the front end later.
Acceptance criteria
1. All code runs locally with a requirements.txt and concise setup notes.
2. Rephrased but semantically similar questions still locate the correct row(s) in the database.
3. Average response time under two seconds on a table of roughly 50 000 rows.
Once this baseline is working, we can discuss optional refinements such as confidence scoring, answer summarisation, or a caching layer, but the first milestone is a reliable question-answer flow backed by the database. Let me know which NLP tools you prefer, any pretrained embeddings you’ll need, and a rough delivery timeline.