spinf Inc.
Spinf provides decision inference at scale: it reads content once and returns calibrated probabilities for many questions in one pass. Its mission is to put leading open-weights language models to work at hyperscale so analyses that were previously too slow or expensive become routine.
At a Glance
- Quantitative research and financial-market intelligence
- Financial crime, compliance and risk teams
- Trust-and-safety teams
- Customer support and operations teams
- +3 more
AI Tools by spinf Inc.
(1)
spinf
Bulk LLM Classification API
Discussions
No discussions yet
Be the first to start a discussion about spinf Inc.
Latest News
Spinf publishes Thinking Text success story: five years of news read from thousands of perspectives, with 15 million articles processed and 92% directional accuracy on large crude moves.
Spinf launches spinf-serve as an Apache 2.0 open-source inference engine for decision models.
Spinf announces the public ghcr.io/spinfinc/spinf-serve container image for GPU deployment.
Spinf publishes official open prompt packs for moderation, triage, product listings, reviews, entertainment metadata and brand perception.
Products & Services
Hosted decision-model API that reads content once and scores hundreds of question/answer options with probabilities, residuals and empty-content floors. It supports on-demand batch work, reserved real-time endpoints, file-based batch processing and dedicated deployments.
Apache 2.0 open-source engine and OpenAI-compatible server for bulk classification and decision inference on one GPU; it supports scoring, extraction, generation slots, media and prompt packs.
Open, ready-made calibrated prompt packs for content moderation, inbound triage, product listings, reviews, entertainment metadata and brand perception.
Python client for the hosted API or a self-hosted spinf-serve instance.
Market Position
Spinf positions itself against conventional general-purpose LLM APIs and chat-oriented inference servers. Its differentiation is decision mode: one content read produces many calibrated, reproducible probabilities, avoiding repeated context processing and generated-text parsing; the hosted service is paired with an open-source, self-hostable engine. This targets high-volume multi-question classification and decision workloads rather than conversational generation.
Leadership
Founders
Giampiero Recco
Founder and CEO of Spinf. Previously founded and ran Beeweeb in Italy for about 12 years, growing it to roughly 80 people before an acquisition; later founded X2TV in San Francisco and raised $4.5 million. His public profile describes him as a VP Engineering/CTO and three-time founder focused on data, AI and platforms.
Executive Team
Giampiero Recco
Founder and CEO
Data, AI and platform executive; previously founded Beeweeb and X2TV. He is identified publicly as the person who designed and built Spinf's full stack and turned its model-optimization and calibration research into a billed self-serve service.
Founding Story
Spinf says it was built initially for its own work: reading millions of news articles each day and asking dozens of calibrated questions per article to produce market indexes. Existing chat-oriented inference servers repeatedly reread the same content for every question and generated answers one token at a time, making the work slow and expensive; Spinf built a one-read/many-decisions engine and published it because people asked to run it themselves.
Business Model
Revenue Model
Usage-based API pricing per million billed input tokens; the content and question tokens are billed, while scores/output tokens are free. Spinf also offers reserved real-time capacity, file-based batch processing and dedicated/private deployments priced per project, while the self-hosted engine is open source.
Pricing Tiers
Launch model; first 50 million tokens are free via $4.50 of starting credit; minimum billing is 1,000 processed tokens per call.
Gemma 4 31B optimized for harder questions; contact sales to enable it.
Runs on the customer's own infrastructure; a 24 GB NVIDIA GPU is the stated minimum for the 12B server.
Target Markets
- Quantitative research and financial-market intelligence
- Financial crime, compliance and risk teams
- Trust-and-safety teams
- Customer support and operations teams
- E-commerce, marketplaces and advertising teams
- Consumer research, brand and social-listening teams
- News analysis and text-as-data research
- Survey and review analysis
- Product, listing and advertising classification
- Content moderation and trust and safety
- Ticket and message intent, urgency and queue routing
- Prompt-injection detection
- Thinking Text