Talent Matched: AI recruitment that scores, screens and sources candidates
ScalaCode built Talent Matched's AI recruitment platform. It has two products, a multi-tenant ATS and a candidate search engine called TM Discover, and both share one AI layer. Their AI agent, Tallie, gets to recruiter data only through MCP servers, and it runs on a private LLM hosted on a dedicated GPU server.
Recruitment intelligence for agencies and hiring teams
Talent Matched builds AI software for recruitment agencies and in-house hiring teams. Its products qualify inbound applications, bring forgotten candidates back from a recruiter's own database and search 850 million external profiles from a plain-English brief. Recruiters work with Tallie, a conversational assistant that explains why each candidate ranks where they do.
What the platform had to do
Search by meaning across a huge talent pool
A recruiter describes the role in plain English. The search then has to find people whose career path and context fit, even when their job titles and keywords don't match the brief.
Run the language model privately
Candidate data stays with Talent Matched. That meant the LLM, the embeddings and the search indexes all had to run on infrastructure the company controls.
One agent across two products
Tallie needed to answer questions and take actions in both the ATS and TM Discover, with no direct access to either database.
Explain every ranking
Before picking up the phone, a recruiter wants to know why a candidate is ranked first.
Keep every customer's data separate
It's a multi-tenant SaaS, and each agency's candidates, notes and pipelines have to stay isolated from everyone else's.
Two products on one private AI layer
There are two products, a multi-tenant ATS and a candidate search engine, and both connect to one AI agent running on a private LLM.
- 01
Multi-tenant ATS
Every customer gets an isolated database schema for its candidates, notes and pipelines. Background workers process CVs and imports, and each employer has a branded careers site.
- 02
TM Discover search engine
Search mixes keyword and semantic matching, and it comes as a web app and a browser extension. Recruiters search from a brief and pull in contact details when they need them.
- 03
Tallie, the AI agent
Tallie only reads and changes data through the MCP servers each product hosts. It remembers the conversation and streams its answers into both apps.
- 04
Private LLM and embeddings
An open-weight language model and an embedding model run on a dedicated GPU server, and prompts and candidate data never go to a third-party model API.
- 05
Vector and graph retrieval
Candidate embeddings sit in a vector database for semantic matching. For questions about relationships, the agent asks a graph database.
- 06
Data ingestion
Pipelines load applications, CVs and old CRM records into the search index, the vector store and the graph.
- 07
Voice screening and tracing
Speech recognition records spoken screening answers so they can be scored. Every agent step and model call is traced, which lets the team debug an answer and check quality.
A plain-English brief in, an explained shortlist out
Recruiters describe a role, refine it in conversation with Tallie and get a ranked shortlist with the reasoning for each candidate. Inbound applications are qualified automatically, old CRM records come back into live roles, and every agent step runs on Talent Matched's own model.
LLM on a dedicated GPU server
sharing one AI agent
What Talent Matched runs on
React web apps, careers sites, browser extension
Python services and background workers
Agent framework with MCP servers and full tracing
Open-weight LLM, embeddings and speech recognition on a private GPU server
Hybrid search, a vector database and a graph database
AWS containers and a dedicated GPU server
Plan a build like Talent Matched for your business
- 30 minutes with the ScalaCode team
- What is realistic, what it costs and how long it takes
- You decide afterwards whether to build with us
