RakshIQ

Self-hosted PII detection & redaction for AI pipelines

5 followers

Self-hosted PII detection & redaction for AI pipelines

5 followers

Rakshiq detects & redacts PII before it reaches your model self-hosted, so data never leaves your infrastructure. What's different: • 100% offline: no phone-home, no data sent to us, ever • Presidio-compatible API a drop-in for existing setups • Built for India: Aadhaar, PAN, GSTIN, IFSC, UPI & more, with checksum validation • Multilingual name/place detection (Hindi + English) • Checksums + libphonenumber, not just regex fewer false positives

Free Options

Launch tags:Privacy•Developer Tools•Artificial Intelligence

Launch Team

Viktor.comAn AI coworker that actually does the work

Promoted

Maker

📌

Hey Product Hunt 👋 I built Rakshiq after watching teams pipe customer data straight into LLMs and third-party APIs with no real idea how much PII was going along for the ride. The usual answer Microsoft Presidio is solid, but it's a Python library you have to host, tune, and babysit, and its coverage of Indian identifiers (Aadhaar, PAN, GSTIN, UPI…) is thin. So Rakshiq started as a simple idea: Presidio-compatible, but self-hosted and with India built in. It keeps a compatible API so existing setups can switch with almost no code change, but everything runs inside your own infrastructure no phone-home, no data sent to us, ever. You can pull the network cable and it still works. The hardest part was false positives. Naive regex flags every 12-digit number as an Aadhaar and every long number as a phone. So detection is backed by real validation Verhoeff and Luhn checksums, libphonenumber, PAN holder-type checks which got India recall to ~97% while keeping precision high. It also handles Hindi and English names/places out of the box. It ships as a Docker image and the license is a signed key you verify offline. Would love your feedback especially which other identifiers or languages you'd want covered next.

Report

2mo ago

Finally someone handling Aadhaar and PAN properly with checksum validation, so many tools miss that and throw false positives everywhere. The Presidio-compatible API was a nice touch, made swapping it into our existing pipeline painless.

Report

2mo ago

Maker

@icguzar82979 Hamdi, thank you! This means a lot - this was exactly why I built RakshIQ solo. I was frustrated with tools just doing bare regex for Aadhaar/PAN and throwing 100s of false positives, so we added real Verhoeff checksum for Aadhaar and holder-type code check for PAN. And yes, kept it Presidio-compatible intentionally so you can just point existing clients to a new base URL. Really happy the swap was painless for you! Would love to hear what you're building with it.

Report

2mo ago

Honestly, the multilingual name detection sounds really useful. One thing I’d love to see added is a simple audit log that shows exactly which fields were redacted and when, just for compliance reviews. That kind of transparency would make it much easier to trust the system in production.

Report

2mo ago

Maker

@leventsaje Hey Levent, thanks so much for this - really good point on trust in production! We actually already have this built in: GET /audit-log records entity types + action taken per call (e.g. PERSON:replace, IN_AADHAAR:mask) with timestamp, but never the raw sensitive text itself. And GET /stats gives you request counts + latency. It was built for multi-tenant / compliance reviews exactly like you described. The dashboard also shows the same per-request. I'll make it more prominent in the docs - thanks for calling this out! Let me know what languages you'd like next, Tamil/Telugu/Bengali are next on our list.

Report

2mo ago

Pros

Cons

Reviews