nchhaya/lora_model
Updated
Niyati Chhaya PRO
nchhaya
AI & ML interests
Document AI, Finance AI , Agentic AI, Document Intelligence
Recent Activity
posted an update 12 days ago
**The hardest part of agentic AI in finance may not be the agent. It may be the document boundary.**
Bank statements are one of the highest-volume documents in finance and one of the least standardized. Every bank has its own layout, columns move, descriptions wrap across lines, and a misread cell can change a lending decision.
Here's what one call to **HyperAPI** does with a real (redacted) statement:
- **Parse** returns layout-aware text and the transaction table, with bounding boxes
- **Extract** returns account-level entities plus 24 line items as clean JSON
- **Redact** masks the account number, customer name and address before anything leaves the pipeline
```bash
curl -X POST https://apis.hyperbots.com/v1/parse \
-H "X-API-Key: $KEY" \
-F "file=@statement.pdf"
```
**What's still hard**, and what we're working on: skewed or low-quality scans, statements where the opening balance sits outside the table, multi-currency accounts, and bundles with several documents in one PDF.
We're building this as the document layer under financial agents, on a simple belief: the agent can be perfect and the outcome still wrong if the input data is wrong.
If you work on document pipelines, I'd like to know what breaks for you. Try it in the playground at https://apis.hyperbots.com and tell me where it falls over.
#documentai #agents #fintech #informationextraction liked a model 13 days ago
google/gemma-4-31B-it liked a model 13 days ago
unsloth/Qwen3.8-27B-GGUFOrganizations
