API Technical Specs
Overview
Sovren offers a resume parsing option that seamlessly integrates the power of Large Language Models (LLMs) with Sovren technology. The result is a parser that is better than can be built with each technology individually.
The LLM Parser can be enabled on a per-request basis. An add-on cost applies to each transaction that has the LLM Parser enabled.
Limitations and Caveats
- The LLM Parser is currently only available for English resumes.
- All fields mentioned in the low-usage fields document are not output. Additionally, the following fields are not output:
- Individual Education Raw Text
- Individual Position Descriptions
- Responses can be several seconds slower when using the LLM Parser.
- The LLM Parser can generate information that is factually incorrect or misleading. It may also output plausible-sounding but false information. However, the LLM Parser offers improved accuracy compared to the standard parser, helping to offset any potential hallucinations.
- In exceptional cases, the LLM Parser returns an error or takes more than 60 seconds to respond. In such cases, we automatically fall back to our standard parser. The add-on credits for using the LLM parser will still be applied.
Frequently Asked Questions
Check our LLM Parser FAQ
FlexRequests (Beta)
Powered by our LLM Parsing Engine, all Resume Parsing requests can include FlexRequests to answer any question about a resume. FlexRequests are available with both the standard, non-LLM parsing engine, and the LLM parsing engine. Each FlexRequest needs to include a Prompt, a DataType, and an Identifier. The possible DataTypes can be found here.
FlexRequests can be set on a per-request basis. An add-on cost applies to each transaction that uses FlexRequests.
Example FlexRequests
"FlexRequests": [
{
"Prompt": "This person has experience working internationally.",
"Identifier": "InternationalWork",
"DataType": "bool"
},
{
"Prompt": "Based on this person's most recent job description, infer the two digit NAICS code for the industry this person is currently working in.",
"Identifier": "Industry",
"DataType": "numeric"
},
{
"Prompt": "How many years has this person spent in a leadership position?",
"Identifier": "YearsLeadership",
"DataType": "numeric"
},
{
"Prompt": "What online courses or trainings has this person completed?",
"Identifier": "OnlineCourses",
"DataType": "list"
},
{
"Prompt": "Is this person a better fit for a senior or junior level position?",
"Identifier": "PositionLevel",
"DataType": "enumeration",
"EnumerationValues": ["Senior", "Junior"]
}
]
Example Replies
"FlexResponse": {
...,
"FlexRequests": [
{
"Identifier": "InternationalWork",
"Reply": "true"
},
{
"Identifier": "Industry",
"Reply": "54"
},
{
"Identifier": "YearsLeadership",
"Reply": "10"
},
{
"Identifier": "OnlineCourses",
"ReplyList": [
"Sun Secure Global Desktop (Tarantella) System Administration"
]
},
{
"Identifier": "PositionLevel",
"Reply": "Senior"
}
]
}
Limitations and Caveats
- FlexRequests are currently only available for English resumes.
- Responses can be several seconds slower when using FlexRequests.
- FlexRequests are powered by Large Language Models (LLMs). LLMs can generate information that is factually incorrect or misleading. It may also output plausible-sounding but false information.