agree. looking to avoid that model.
no incentive for investors in staying, purely a trading/yield play.
5 karma · joined November 27, 2019
https://useorin.com/blog/cheat-sheet-for-enabling-ai-agents-...
Feel free to take a look and suggest any feedback.
-Cheers
Fintech AI Agents can now perform cascaded tasks using multiple underlying foundation models. Delivering the best in performance, agentic actions, and execution costs.
A few highlights:
▶ We’ve added Deepseek R1 as one of the foundational models for Orin AI agents. Seeing pretty solid initial results for logical reasoning when tested on a few B2C fintech scenarios. *Currently #Deepseek is available only in private beta (without SOC-2).
▶ Our customers and prospects have been asking about Claude. As it excels in security and stepwise actions — both critical for fintech use cases. So, we’ve included support for Claude Sonnet 3.5 for Orin’s Agentic actions.
▶ Orin now supports three modes of customer support engagement: 1/ User chat – Embedded on customer portals and help center sites. Answers product questions and resolving Tier-0 issues.
2/ Email queries – Parses email threads to understand customer problems, looks up different sources, reasons out a solution, and drafts responses. Used by human agents to reply to customers efficiently.
3/ Ticket forms – Assists human agents by analyzing support tickets, using macros, knowledge bases, and past history. Suggest responses for customer reps handling the issue.
Payment apps – Handling a variety of user inquiries on rates, payment methods, transaction times, and currency exchange. Wealth mgmt & Tax accounting – Managing queries about IRS forms, tax limits, 401(k), and IRA options.
An example AI Agent looks like this inside the customer portal --> See the full blog in URL.
And did the same on ChatGPT UI and OpenAI API to test answers, for a side-by-side comparison.
Highlights
Answer Quality I liked the reasoning aspect of Deepseek R1, which explains the logical details behind the answering. It helps in fine-tuning the AI agents and debugging for better responses. This option is easily collapsible, so we can show/hide it to end users.
I used the distilled model which is resource-efficient and has limited reasoning abilities. But the full model can give us: ▶ Multi-step problem-solving ▶ Better reasoning ▶ Self-verification capabilities
Speed Running this on my local machine (Apple M1 Pro 16GB), even using the smallest model - R1.b slowed everything on my system. Development on local 16GB is not optimal for Gen AI training and inference, so I understand. So this is easily solvable when we move to cloud hosting on AWS or GCP instances. As well use Deepseek's bigger models.
Cost It is free. We don’t have to worry about tokens during training and inference. When deployed for application AI models, this can help us drastically bring down the cost of production and subscription fees.
Security I have not dug much into this part yet, but soon will. Given B2B and Fintech use cases require tight customer data security and privacy, this is critical.
Currently, Orin uses hashtag#OpenAPI for Fintech AI agents apps, but anything with customer deployment using hashtag#Deepseek will need to be scrutinized for many months, passing SOC2 and other compliance standards.
Setup Please refer to the blog post in link
In this solution requirement <> vendor matrix, you'll find categorized items of - Features / UI needed for customer support use cases - Reporting and analytics for business KPIs - Data and integration to customer support software - Administration and Security systems for team management
This document is aimed at managers and directors looking to evaluate different AI-based solutions available in market and help improve productivity of their teams.
If you're currently looking for AI solution, I'm happy to share the document and discuss how can we improve evaluation? Ping me in this thread or DM me.
IRL events are a great way to connect to the ecosystem of founders, investors, and customers. Most events are free, but finding and reminding about them are hard. We put the best startup events straight into your calendar.
What do you like to see here? Please provide us with feedback.