ApplyLogic The challenge
The UK placement-year hunt is punishing. Roles are scattered across job boards, company scheme pages and university portals; the same opening shows up in five places with slightly different details; deadlines come early and don’t wait; and a student carrying a full course load is expected to trawl, cross-reference and tailor applications to dozens of postings — with little real sense of which ones actually fit. Most end up applying to whatever’s most visible and quietly giving up on the long tail, missing roles that would have suited them better.
It’s exactly the kind of high-volume, repetitive, judgement-heavy work that grinds down a person doing it by hand — and exactly the kind of work a well-designed system of AI agents can take on.
The solution
Fledge is built around a team of specialised AI agents that, between them, do the job of a tireless research assistant: continuously find placement roles, make sense of them, and match them to the individual student. And ApplyLogic didn’t just design these agents on a whiteboard — it built them, hardened them and deployed them into production, where they run autonomously around the clock.
A multi-agent pipeline
Rather than lean on one monolithic prompt, Fledge coordinates a set of focused agents, each with a clear remit — discovery agents that continually scan sources for new placement roles; extraction agents that read messy, inconsistent listings and turn them into structured, comparable data; a deduplication step that recognises when the same role appears across multiple sources and collapses it into one; and matching agents that weigh each role against a student’s profile. They’re orchestrated and run on schedules, so the picture stays current without anyone lifting a finger.
Understanding fit, not just keywords
Because the matching layer is driven by a language model, it reasons about fit the way a good careers adviser would — past keyword overlap to the substance of the role and the student’s strengths, course and goals. Each student gets a ranked shortlist, not a haystack to dig through.
Clean data students can trust
A job-hunt tool is only as good as its data, so a real share of the engineering went into deduplication and normalisation — making sure a student is never shown the same role five times or asked to compare listings that don’t line up. Trustworthy data is what makes the shortlist worth acting on.
Safety where agents touch real systems
Agents that can read and write to live data need guardrails. Fledge wraps its agents in safety layers so autonomous steps stay within well-defined bounds — the discipline that separates a dependable agentic product from an impressive demo.
Event-driven and correct under pressure
Underneath, an event-driven pipeline of Go services over NATS JetStream, backed by Neon Postgres, moves each role through discovery, extraction, deduplication and scoring. Processing is idempotent, so even a redelivered message can never corrupt a student’s match results — the kind of correctness that matters enormously once agents are running continuously and unattended.
The results
- Hours of manual trawling replaced by a curated, ranked shortlist — students see the roles that fit, first.
- Wider coverage — placements surfaced that a manual search would simply never reach.
- Always current — the agent pipeline runs continuously, so the list is never stale.
- A calmer, less demoralising search — for students already stretched thin.
Why it worked
Fledge is the clearest expression of what ApplyLogic does: building AI-enabled solutions where AI isn’t a feature bolted onto the side, but the engine of the product itself. A coordinated system of agents — built with frontier models and engineered with the safety, idempotency and data discipline that make autonomy trustworthy — takes a problem that overwhelms people and handles it gracefully. Designing, orchestrating and hardening systems like this is the core of ApplyLogic’s work; Fledge is what it looks like end to end.