Token efficient domain recommendations

1 . A method comprising:

receiving a query;

classifying the query to a first domain within a plurality of domains, based at least in part on average computational resource usage across the plurality of domains, comprising:

calculating a token count adjusted confidence score for each domain of the plurality of domains as a quotient of a weighted confidence score of each domain and a function of an average token count of each domain modulated by a tunable hyperparameter;

retrieving an index of domain-specific vector embeddings corresponding to the plurality of domains;

prompting a Large Language Model (LLM) with the query and the index of domain-specific vector embeddings for retrieval augmented generation (RAG) of a query response;

generating, by the LLM, the query response grounded with a relevant domain-specific vector embedding of the index of domain-specific vector embeddings as a contextual anchor, wherein the query response is aligned with a semantic space of the relevant domain-specific vector embedding;

forwarding the query response over a network communication channel to a downstream application; and

displaying, by the downstream application, the query response in a user interface.

2 . The method of claim 1 , further comprising:

calculating a global confidence score; and

comparing the plurality of domains and suggesting the plurality of domains.

3 . The method of claim 1 , wherein the weighted confidence score is calculated as:

weighted

⁢

confidence

j

=

∑

i

=

1

k

(

sigmoid

i

j

×

weight

i

j

)

+

η

j

wherein:

j is a domain,

k is a number of top results,

i is a current index,

sigmoid is a sigmoid function,

weight is a confidence score weight, and

η j is a domain specific biasing constant.

4 . The method of claim 3 , wherein the domain specific biasing constant is calculated as:

η

j

=

ω

c

×

C

j

+

ω

h

×

H

j

+

ω

p

×

P

j

wherein:

ω c is a weight for a current context bias,

C j is a current context relevance to domain j,

ω h is a weight for a conversation history bias,

H j is a conversation history relevance to domain j,

ω p is a weight for a popularity bias, and

P j is a popularity of domain j.

5 . The method of claim 2 , wherein the global confidence score is calculated as:

GCS

=

μ

TCACS

+

β

+

σ

TCACS

wherein:

μ is a mean of TCACS across the plurality of domains,

β is a tunable hyperparameter, and

σ is a standard deviation of TCACS across the plurality of domains.

6 . The method of claim 2 , wherein comparing the plurality of domains further comprises:

if the token count adjusted confidence score of the first domain is greater than the global confidence score, retrieving results from the first domain; and

if the token count adjusted confidence score of the first domain is not greater than the global confidence score, retrieving the results from the plurality of domains.

7 . A system comprising:

a computer processor;

memory; and

instructions stored in the memory and executable by the computer processor to cause the computer processor to perform operations, the operations comprising:

receiving a query,

classifying the query to a first domain within a plurality of domains, based at least in part on average computational resource usage across the plurality of domains, comprising:

calculating a token count adjusted confidence score for each domain of the plurality of domains as a quotient of a weighted confidence score of each domain and a function of an average token count of each domain modulated by a tunable hyperparameter,

retrieving an index of domain-specific vector embeddings corresponding to the plurality of domains,

prompting a large language model with the query and the index of domain-specific vector embeddings for retrieval augmented generation (RAG) of a query response,

generating, by the LLM, the query response grounded with a relevant domain-specific vector embedding of the index of domain-specific vector embeddings as a contextual anchor, wherein the query response is aligned with a semantic space of the relevant domain-specific vector embedding;

forwarding the query response over a network communication channel to a downstream application; and

displaying, by the downstream application, the query response in a user interface.

8 . The system of claim 7 , further comprising:

calculating a global confidence score; and

comparing the plurality of domains and suggesting the plurality of domains.

9 . The system of claim 7 , wherein the weighted confidence score is calculated as:

weighted

⁢

confidence

j

=

∑

i

=

1

k

(

sigmoid

i

j

×

weight

i

j

)

+

η

j

wherein:

j is a domain,

k is a number of top results,

i is a current index,

sigmoid is a sigmoid function,

weight is a confidence score weight, and

η j is a domain specific biasing constant.

10 . The system of claim 9 , wherein the domain specific biasing constant is calculated as:

η

j

=

ω

c

×

C

j

+

ω

h

×

H

j

+

ω

p

×

P

j

wherein:

ω c is a weight for a current context bias,

C j is a current context relevance to domain j,

ω h is a weight for a conversation history bias,

H j is a conversation history relevance to domain j,

ω p is a weight for a popularity bias, and

P j is a popularity of domain j.

11 . The system of claim 8 , wherein the global confidence score is calculated as:

GCS

=

μ

TCACS

+

β

+

σ

TCACS

wherein:

μ is a mean of TCACS across the plurality of domains,

β is a tunable hyperparameter, and

σ is a standard deviation of TCACS across the plurality of domains.

12 . The system of claim 8 , wherein comparing the plurality of domains further comprises:

if the token count adjusted confidence score of the first domain is greater than the global confidence score, retrieving results from the first domain; and

if the token count adjusted confidence score is not greater than the global confidence score, retrieving the results from the plurality of domains.

13 . A computer program product comprising a non-transitory computer-readable medium having stored thereon program code that, when executed by one or more computer processors, causes the one or more computer processors to perform operations comprising:

receiving a query;

classifying the query to a first domain within a plurality of domains, based at least in part on average computational resource usage across the plurality of domains, comprising:

calculating a token count adjusted confidence score for each domain of the plurality of domains as a quotient of a weighted confidence score of each domain and a function of an average token count of each domain modulated by a tunable hyperparameter,

retrieving an index of domain-specific vector embeddings corresponding to the plurality of domains;

prompting a Large Language Model (LLM) with the query and the index of domain-specific vector embeddings for retrieval augmented generation (RAG) of a query response;

generating, by the LLM, the query response grounded with a relevant domain-specific vector embedding of the index of domain-specific vector embeddings as a contextual anchor, wherein the query response is aligned with a semantic space of the relevant domain-specific vector embedding;

forwarding the query response over a network communication channel to a downstream application; and

displaying, by the downstream application, the query response in a user interface.

14 . The computer program product of claim 13 , wherein classifying the query further comprises:

calculating a global confidence score;

comparing the plurality of domains; and

suggesting the plurality of domains.

15 . The computer program product of claim 13 , wherein the weighted confidence score is calculated as:

weighted

⁢

confidence

j

=

∑

i

=

1

k

(

sigmoid

i

j

×

weight

i

j

)

+

η

j

wherein:

j is a domain,

k is a number of top results,

i is a current index,

sigmoid is a sigmoid function,

weight is a confidence score weight, and

η j is a domain specific biasing constant.

16 . The computer program product of claim 15 , wherein the domain specific biasing constant is calculated as:

η

j

=

ω

c

×

C

j

+

ω

h

×

H

j

+

ω

p

×

P

j

wherein:

ω c is a weight for a current context bias,

C j is a current context relevance to domain j,

ω h is a weight for a conversation history bias,

H j is a conversation history relevance to domain j,

ψ p is a weight for a popularity bias, and

P j is a popularity of domain j.

17 . The computer program product of claim 14 , wherein the global confidence score is calculated as:

GCS

=

μ

TCACS

+

β

+

σ

TCACS

wherein:

μ is a mean of TCACS across the plurality of domains,

β is a tunable hyperparameter, and

σ is a standard deviation of TCACS across the plurality of domains.