Intelligence, applied where it's accountable.
Retinex builds applied-intelligence products for document analysis, financial underwriting, and operational workflows, deployed on-premise where data can't leave the building.
Six products. Three in active pilot. One national AI Grand Challenge finalist.
- AI Grand Challenge 2025, for speed, accuracy and versatility
-
Top 3 - Recognised startup, Startup India
-
DPIIT - Products in portfolio
-
6
Recognised by
DPIIT-registered under Startup India. Finalist, national AI Grand Challenge 2025.
The portfolio
Six products and projects across regulated and operational domains.
Each solves a specific, hard problem: some ship as products now in pilot, others as bespoke project builds delivered end to end.
DocFlow
Credit intelligenceDocument-grounded credit underwriting for banks and NBFCs: financials, bank statements, and a behaviour score, source-linked.
DocIntel
Document extractionOn-premise multilingual document extraction: layout, OCR, tables and figures, across many languages.
PCB Inspector
Vision inspectionAutomated optical inspection for electronics lines: detect faults and missing components against a calibrated reference, at production speed.
Vilo
Interior-design SaaSEnd-to-end operations for interior-design firms: from lead to GST-compliant quote to milestone invoice.
Animatronics Timeline
Motion controlA timeline editor for animatronics, and the firmware that drives the rig: keyframe each motor, scrub to preview, then run it on the hardware.
DrillVision
Vision-guided automationCamera-guided drilling for circuit boards: find each target on the board and place the drill on it precisely, panel after panel.
AI Grand Challenge 2025
Selected to the national top three.
Our on-premise document-extraction system was chosen as a finalist in India's AI Grand Challenge, selected for its speed, accuracy, and versatility across a demanding multilingual brief.
The brief was demanding: extract structure, text, tables and figures from documents across many languages, right-to-left and Devanagari scripts included, running entirely on-premise. We built it, and it runs on a single node, behind your firewall.
- Multilingual
- Tables extracted
- Figures captioned
- Single-node deploy
AI Grand Challenge 2025-26
Official Stage 2 results
Problem Statement 5
Intelligent Multi-lingual Document Understanding
Progression · PS-5
Pre-Evaluation
15 teams
Stage 1
Top 6
Stage 2
Top 3
- Retinex Technologies Pvt. Ltd. Our team
- Finalist team, name withheld withheld
- Finalist team, name withheld withheld
Official Stage 2 shortlist for PS-5: top 3 nationally. The published order is not the ranking, and the other two team names are withheld out of professional courtesy.
Source: ai-grand-challenge.in
Recognised by the Government of India
A recognised startup.
Retinex is recognised as a startup by the Department for Promotion of Industry and Internal Trade (DPIIT) under the Government of India's Startup India initiative, on record as working in artificial intelligence.
It is the same standing referenced across this site: a verifiable, government-issued credential that sits behind the products, not a self-applied label.
- Certificate
- DIPP108856
- Recognised by
- DPIIT, Government of India
- Industry
- Artificial Intelligence
- Recognised since
- 2022, valid to 2030
How it's built
Depth where it's hard to fake.
Our products share a foundation: document understanding, grounded reasoning, and deployment that respects where regulated data has to live.
Document intelligence
Layout detection, OCR across many languages including right-to-left and Devanagari scripts, and table and figure extraction to structured JSON, XML, and Markdown.
output: json · xml · md · html Grounded analytics
Every figure traces back to its source. Answers in our credit workspace link to the exact page and line they came from, with no unsourced claims.
every answer → source-linked On-premise deployment
Packaged to run inside your own infrastructure on a single GPU node, so no document leaves your network.
docker · single-node · gpu Engineered end to end
Custom-trained models served on GPU, real-time computer vision on the line, and typed, maintainable application code: one foundation across every product.
custom-trained models · gpu inference · computer vision Areas of expertise
The applied-AI foundations our products are built on.
-
Large language models (LLMs)
Reasoning and extraction over natural-language documents.
-
Vision-language models (VLMs)
Joint understanding of images and text.
-
Computer vision
Detection and inspection on visual data.
-
Machine learning
Models trained for the task, not off the shelf.
-
Grounded reasoning & RAG
Answers tied back to their source.
-
AI application hosting & scaling
Serving models on-premise, scaled to load.
Why now
Regulated and operational economies are digitising, and they need software that can be trusted with the decisions that matter.
Across industries, work that carries real consequences (an application form turned into data, a circuit board cleared to ship) is moving off paper and manual checks. But it is unforgiving: close enough is not enough, every result must be explainable, and none of it can leave the building.
That is the gap Retinex is built for: domain-specific products that read the documents, watch the line, or drive the machine, each one accurate, traceable, and on-premise. Three are in pilot, we placed in India's flagship deep-tech challenge, and the next phase is scale.
“We don't sell automation. We sell work that can be audited.”
Work with us
Applied intelligence, built for regulated India.
We're raising to scale three pilots into deployments. Request the investor deck, or book a walkthrough of any product in the suite.