NanoNets has rolled out OCR-3, a 35-billion parameter Mixture-of-Experts visual language model designed specifically for agents and retrieval-augmented generation workloads. According to a release distributed on Issuewire, the model topped public optical character recognition benchmarks, scoring 93.1 on OLM-OCR and 85.9 on the IDP Leaderboard. The model was trained on 11 million documents and runs at twice the speed of its predecessor, OCR-2.
The company is using the release to anchor a technical push into autonomous back-office processing. Rather than outputting bare text, OCR-3 introduces native visual question-answering along with bounding-box spatial coordinates and itemized confidence scores for each extraction. To guide the next generation of these architectures, NanoNets is looking for a VP of AI ($240,000–$300,000) in Palo Alto, CA, US / Remote (Palo Alto, CA, US). The position oversees fine-tuning state-of-the-art vision-language models for workflow automation and constraining generative architectures to eliminate hallucinations while outputting structured workflows for invoices, receipts, and passports.
Automating Finance Workflows
Founded in 2017, NanoNets builds software that converts unstructured files—such as bills of lading, purchase orders, contracts, and tax forms—into structured data and routes them into systems like Salesforce, QuickBooks, and enterprise resource planning platforms. The company reports that over 34% of Global Fortune 500 companies have used its automation platform, with more than 10,000 companies adopting the system across finance, operations, and accounting.
A central focus of the company is straight-through processing (STP), which measures the share of documents processed without human intervention. Standard invoice processing requires around 15 minutes of manual data entry, approvals, and general ledger lookups. NanoNets reduces that time to under one minute, delivering an STP rate above 90% across millions of monthly documents. In an interview with CNBC-TV18, chief executive officer Sarthak Jain explained that reading characters is an older problem, while the operational task centers on context: "how do you make meaning from all of the information that's present on this piece of paper and should you pay this vendor the amount that they are asking for that's the real question."
The Iteration to Document Processing
NanoNets was started by Sarthak Jain and chief technology officer Prathamesh Juvatkar. Both graduated with electrical engineering degrees—Jain from the Indian Institute of Technology Madras and Juvatkar from the Indian Institute of Technology Gandhinagar. Prior to NanoNets, the pair co-founded Cubeit, which was acquired by Myntra.
After entering the Y Combinator W17 batch, the team experimented with visual recognition tools for drone imagery and content moderation. In a conversation hosted by Elevation Capital, Jain recounted that the business initially reached $1.5 million in annual recurring revenue across divergent custom use cases before resetting to zero to focus purely on a scalable, repeatable document extraction product.
That pivot established the basis for institutional rounds. NanoNets raised a $10 million Series A led by Elevation Capital in 2022, followed by a $29 million Series B led by Accel, bringing total equity funding to $42 million. Additional investors include SV Angel, Sound Ventures, and Soma Capital.
Engineering and Distributed Training Stack
The company's technical stack relies on Golang for its external APIs and microservices, paired with Python, TensorFlow, and PyTorch for machine learning pipelines. The backend team compiles Python into C to export binaries into Golang for on-premise installations, while workloads scale across Kubernetes on Amazon Web Services and Google Cloud Platform using custom GPU metrics.
To build out training capacity, the engineering team in India is hiring a Senior Deep Learning Engineer in Bengaluru, KA, IN / Bengaluru, Karnataka, IN. The role entails training transformer-based models and custom vision-language models for document understanding at scale, as well as designing distributed training pipelines for billion-parameter architectures using PyTorch Fully Sharded Data Parallel (FSDP) and DeepSpeed.
Go-to-Market Growth in the United States
Around 75% of company revenue originates from North America and Europe. To support inbound enterprise demand and post-sale onboarding, NanoNets has opened several customer-facing roles centered around its Palo Alto location.
Commercial additions include an Account Executive - US ($100,000–$120,000) working in a hybrid arrangement in Palo Alto to close inbound enterprise leads, and a Sales Engineer ($90,000–$140,000) to build custom machine learning models on customer data for technical sales demos. For positioning, the team is recruiting a Product Marketing Manager ($150,000–$200,000) across Bay Area locations to produce sales enablement assets and pipeline messaging.
Post-sale teams are also adding capacity. An Engagement Manager ($160,000–$180,000) in Palo Alto guides customer business outcomes, while an Account Manager (Customer Success) ($80,000–$140,000) manages client expectations, net revenue retention, and up to 15% travel.
Open Roles at NanoNets
| Role | Location | Work Mode | Compensation |
|---|---|---|---|
| VP of AI | Palo Alto, CA, US / Remote (Palo Alto, CA, US) | Remote | $240,000–$300,000 USD |
| Account Executive - US | Palo Alto, CA, US / Remote (Palo Alto, CA, US) | Remote | $100,000–$120,000 USD |
| Senior Deep Learning Engineer | Bengaluru, KA, IN / Bengaluru, Karnataka, IN | Onsite | Not listed |
| Sales Engineer | Palo Alto, CA, US | Onsite | $90,000–$140,000 USD |
| Product Marketing Manager | Palo Alto, CA, US / San Francisco, CA, US / San Jose, CA, US / Menlo Park, CA, US / Mountain View, CA, US / Sunnyvale, CA, US | Onsite | $150,000–$200,000 USD |
| Engagement Manager | Palo Alto, CA, US | Onsite | $160,000–$180,000 USD |
| Account Manager (Customer Success) | Palo Alto, CA, US / Remote (San Francisco, CA, US) | Remote | $80,000–$140,000 USD |