Triniti APIs empowers financial institutions with cutting-edge capabilities to offer their customers a natural, personalised, intelligent and low-cost channel of engagement via voice and text.
Triniti’s state-of-the-art AI algorithms enables it to identify and understand intent, interpret discourse and reactions, extract entities and recognize values, analyse sentiments and emotions in conversations and much more, transforming access to financial services to be more natural, seamless and invisible.
Our ability to craft intelligent micro conversations is due to the fact that we tuned our AI engine specifically for banking and finance and pre-trained relevant ontology that is build with preprocessor, NLP, NLU, NLG and Machine Comprehension in mind.
Corrects misspelt words based on user’s input. Knows not to correct names, nicknames etc. Expand and normalises acronyms like “acc” to account, “xfer” to transfer. Able to standardise forms like “CreditCard” to Credit Card.
Advanced algorithm to convert “Yesterday”, “1st Sunday of next month”, “last 4 weeks”, etc to standardised date formats.
Convert numbers expressed in words like “five hundred and twenty thousand” into numerical form 520000.
Can detect up to two instructions in a sentence and separate them to inherit required entities. For example, “Pay 500 to Jack and Jill” becomes “pay 500 to Jack” and “pay 500 to Jill”.
Detects if the user’s utterance is in context of the current conversation. Also detects implied cancellations, confirmations and negation requests.
Able to detect if the user’s current input is positive, neutral or negative. Detects if the user is being abusive and able to respond appropriately.
Detects and provides options when user’s input is a fragment and is not expandable. For example, “Show..” could mean “show balance, show due bills, show transactions”.
Auto Calibrating Classifier that is tolerant to differences in instances of training data across intents, able to differentiate between FAQ and non-FAQ intelligently and detect adversity.
Highly Trainable with trainable entity linking and Auto Calibrating features. Able to extract entities with high precision and confidence. Extracts and links Modifiers like "from" and “to” to related entities. Able to extract modifiers that modify the complete intent, such as “minimum” in “Minimum Balance”.
Provides the response type (“Numeric”) that the user is expecting and its grain (“Count”). It helps bots respond with more precise answer than giving generic response.
Detects if the sentence is in past tense or not, in order to help bots interpret the request correctly. For example, “How many EMIs have I paid” vs “How many EMIs do I have to pay”.
Extracts relationship between entities thus enabling bots to handle much more complex requests with multiple entities, constraints, and more.
Responds to most commonly-used ‘small-talk’ or ‘chit-chat’ that users have with bots.
Trained with a large corpus of Q&As with variants, providing direct responses to general product/process related queries from the user.
Guide the user with related queries or transactions when user’s input is not clear, or whenever appropriate, to enhance user’s experience.