99.9% availability and sub-100ms responses for a recruitment intelligence platform, at negligible infrastructure cost
Aggregates job listings and candidate data from across the web and matches candidates to roles in both directions, with the hiring contact attached.
- Recruitment technology
- Cloud and platform
- Applied AI
- Product engineering
- <100 ms
- API response time at p95
- 99.9%+
- Availability
- <5 s
- AI processing time for complex media
- Usage-priced
- Infrastructure that costs almost nothing when idle
In brief
- Hexploits designed and built JobVantage’s recruitment intelligence platform end to end on usage-priced AWS services so that infrastructure cost tracks usage.
- The platform ingests CVs, voice notes, screenshots and documents, extracts structured candidate profiles with confidence scores, and matches in both directions.
- API responses are under 100 milliseconds at the 95th percentile with 99.9% or better availability; complex media is processed in under five seconds.
- Each match is scored out of 1,000 with written reasoning and the hiring contact attached; the relationship is ongoing.
Client context
JobVantage sells recruitment intelligence to agencies. Its platform aggregates job listings and candidate data from specialist providers, extracts structure from whatever a recruiter has to hand, and ranks matches in both directions: candidates for a role, and roles for a candidate, with the hiring contact attached so the recruiter can act immediately.
It is a growing UK technology business selling to recruitment agencies, and its costs had to stay in proportion to its revenue.
The situation before
JobVantage came to us with a specification and no platform. The constraints were clear:
- Load is unpredictable. Job postings spike on Monday mornings and candidate activity surges after redundancy announcements. Paying for idle capacity between spikes was not acceptable.
- Candidate information arrives as CVs in every layout, voice notes from calls, screenshots of portfolios and documents in dozens of formats. All of it had to become structured, searchable data without manual entry.
- Data comes from several third-party providers, each with its own API, rate limits and formats. New sources had to be easy to add.
- Matching had to work both ways, with a ranking a recruiter can trust and explain.
- Infrastructure cost had to be close to zero until there was revenue to pay for it.
What we did
We designed and built the platform end to end on usage-priced cloud services, so the bill rises and falls with the work being done and there is nothing to keep running when the platform is quiet.
We built the ingestion pipeline that takes documents, audio and images and produces structured candidate profiles with confidence scores and a record of where each fact came from. We built the integration layer that hides each data provider's quirks behind one model, monitors their health and routes around outages. We built the matching engine: semantic search across the whole corpus, then a ranking stage that scores each match out of 1,000 with written reasoning, and returns the thirty best with the decision-maker's contact details. And we built the APIs agencies use to embed all of this in their own systems.
The relationship is ongoing.
Results
| Measure | Result |
|---|---|
| API response time | Under 100 ms at the 95th percentile |
| Availability | 99.9% or better |
| AI processing time for complex media | Under 5 seconds |
| Monthly infrastructure cost | Rounds to almost nothing at current usage; scales linearly with load |
| Matching output at delivery | Top 30 ranked matches per query, scored out of 1,000, with reasoning and hiring contact |
What the client said
Working with Hexploits has genuinely been a pleasure, and I see them as my scaling partner for the foreseeable future as JobVantage grows. If you're looking for a development team who combine strong AI/engineering capability with honesty, flexibility and a real interest in your business, I'd strongly recommend them.
Brandon Bowden, Founder, JobVantage
For the IT reader: technical detail
Compute and data. All backend compute runs on AWS Lambda with sub-100 ms cold starts; DynamoDB in on-demand mode provides single-digit millisecond reads with global secondary indexes for both job and candidate query patterns; SQS decouples ingestion from AI processing with dead-letter queues for retries. Each function scales independently, so a spike in image processing does not affect API latency.
Why serverless rather than containers. Container platforms scale reactively and need scaling policies, node pools and an orchestration layer to run, which is engineering time or operational risk for a business without a platform team. Usage-priced services scale automatically and cost near zero when idle. The trade-off is accepted deliberately and documented.
Multi-modal extraction. PDF and Word parsing with OCR for scans; speech-to-text with speaker diarisation for recordings; vision models for screenshots; schema mapping for structured exports. Large language models parse CVs regardless of layout and reconcile conflicting information across sources. Each profile carries normalised skills and experience, semantic embeddings, per-field confidence and source provenance.
Integrations. A provider abstraction with unified data models, queue-based ingestion with backoff for rate-limited sources, health monitoring, automatic failover and a normalisation layer for differing skills taxonomies.
Matching. Vector similarity search narrows millions of candidates or roles to a focused set; a model-based ranking stage scores each out of 1,000 with reasoning and confidence indicators; contact enrichment attaches the hiring manager or recruiter with LinkedIn, email and phone where available. Jobs, Candidates, Companies, Matching and Contacts APIs are RESTful with authentication, rate limiting and webhooks.
Frontend. Deployed on ECS Fargate behind an Application Load Balancer with health checks, managed TLS and rolling zero-downtime releases.
Questions about this engagement
Why serverless rather than containers?
How does the matching work?
What data sources does it use?
Related
More in Recruitment and HR technology.
Gradvisor · Charity · 8 weeks to production, then ongoing
98% faster page loads and a 12% smaller cloud bill for a UK careers charity
swarmd.ai · Software vendor · 6 months
Enterprise AI control plane delivered in six months at 75% under budget for a UK software vendor
Next step
Have a system like this one?
Tell us about the system and the sector. A named engineer replies within one working day. A written scope and an indicative price within two working days of a short scoping call.