You will get AI-powered Text-to-SQL system translating NL questions into SQL queries
Project details
RAG significantly improved schema accuracy — reduced hallucinated columns
by ~80%.
Fine-tuned Qwen 3 generalized well across similar financial queries.
Ollama simplified local orchestration, eliminating dependency on cloud LLM
APIs.
Challenge: Balancing token length in prompt due to large schema context.
Future Work: Implement query correction feedback via reinforcement (RLHF-lite).
This project successfully demonstrated a domain-adapted Text-to-SQL system
leveraging open-source LLMs within a fully local and private environment.
By combining RAG from Llama 3.2 and fine-tuning with Qwen 3, the system achieves
near-enterprise-grade SQL generation accuracy suitable for financial analytics, while
remaining cost-efficient and privacy-preserving.
9. Tools and Technologies
Layer
Tool / Framework
Model Runtime Ollama, Transformers, PEFT
Base Models Llama 3.2, Qwen 3
Database
Vector Store
Evaluation
PostgreSQL / MS SQL
FAISS / Milvus
BLEU, EM Accuracy, SQLGlot
Orchestration LangChain or LlamaIndex
Interface
FastAPI / Streamlit
by ~80%.
Fine-tuned Qwen 3 generalized well across similar financial queries.
Ollama simplified local orchestration, eliminating dependency on cloud LLM
APIs.
Challenge: Balancing token length in prompt due to large schema context.
Future Work: Implement query correction feedback via reinforcement (RLHF-lite).
This project successfully demonstrated a domain-adapted Text-to-SQL system
leveraging open-source LLMs within a fully local and private environment.
By combining RAG from Llama 3.2 and fine-tuning with Qwen 3, the system achieves
near-enterprise-grade SQL generation accuracy suitable for financial analytics, while
remaining cost-efficient and privacy-preserving.
9. Tools and Technologies
Layer
Tool / Framework
Model Runtime Ollama, Transformers, PEFT
Base Models Llama 3.2, Qwen 3
Database
Vector Store
Evaluation
PostgreSQL / MS SQL
FAISS / Milvus
BLEU, EM Accuracy, SQLGlot
Orchestration LangChain or LlamaIndex
Interface
FastAPI / Streamlit
Database Type
MS SQLWhat's included
| Service Tiers |
Starter
$3,500
|
Standard
$8,500
|
Advanced
$11,000
|
|---|---|---|---|
| Delivery Time | 20 days | 60 days | 70 days |
Number of Revisions | 0 | 1 | 2 |
Number of Queries | 0 | 2 | 2 |
Query Debugging | - | - | |
Query Optimization | - | - | |
Query Scheduling | - | - | |
Query Analysis | - | - | |
Source Code | - | - |
Optional add-ons
You can add these on the next page.
Query Optimization
(+ 1 Day)
+$150
Query Scheduling
(+ 1 Day)
+$250
Query Analysis
(+ 1 Day)
+$200About Bandi
Data science AI/ML
Isnapuram, India - 9:06 pm local time
geospatial, supply chain, banking, and finance domains. Proven expertise in Generative AI, RAG pipelines, OCR
systems, Text-to-SQL, Speech Recognition, and ML/DL solutions. Skilled at building end-to-end AI/ML pipelines,
deploying scalable APIs, and delivering enterprise-ready AI applications.
Steps for completing your project
After purchasing the project, send requirements so Bandi can start the project.
Delivery time starts when Bandi receives requirements from you.
Bandi works on your project following the steps below.
Revisions may occur after the delivery date.
Objective --> OBDC --> OLLMA-->Local Docker Setup-->Streamlit App