Torre.
I helped recruiters find candidates and assess their experience.
Quality improvement from the selected model integration after comparing three LLM providers.
Lower cost for the selected integration in the model comparison.
More candidate profiles appeared in Google.
Useful public profiles were missing from search results, while outdated pages stayed listed. I rebuilt the sitemaps and refreshed listings when profiles, companies, or hiring status changed.
Alex Okafor
Software engineer
Public profiles.
Current sitemaps.
sitemap.xmlUpdated/maya-chenPublic/alex-okaforIncluded/jordan-reedPublicUseful URLs for search engines.
A sitemap helps discovery; it does not guarantee indexing.
Profile, company, and careers pages use separate inclusion rules. A content change refreshes the relevant listing.
How I built itRead the engineering details
A longer sitemap could still send a search engine to empty or outdated pages. I focused on useful coverage and kept existing listings current through events and cache invalidation. An update needed to change the relevant listing without rebuilding every directory.
- Scala & SQL
- Build the sitemap endpoints and select the records each directory needs.
- Events & RabbitMQ
- Refresh affected listings when content or hiring activity changes.
- XML sitemaps & cache invalidation
- Serve searchable directories without continuing to return stale versions.
+1,700% Google-indexed pages. Growth in indexed candidate and company pages after the sitemap improvements.
Relevant experience became easier to spot.
General profiles made relevant experience easy to miss. I built profile introductions and résumé views tailored to an opening, with checks that kept qualifications faithful to the candidate’s original experience.
Relevant experience.
Original facts.
Alex Okafor
Software engineer
I build backend APIs01 and background jobs02 for inventory and product data.
Select an experience to see its supporting claim. The relevant work comes forward; qualifications are not invented.
How I built itRead the engineering details
The model could borrow a skill from the job description that the candidate had never claimed. Checks had to preserve titles, figures, and supporting experience. There was also a waiting problem: I used priority queues and concurrency control so a profile being opened could take the next available generation slot.
- Scala & structured LLM output
- Turn model responses into fields the application can check and display.
- Grounding & validation
- Compare generated claims with the person’s existing experience.
- Priority queues & concurrency control
- Give opened profiles precedence over waiting background work while controlling simultaneous generation.
+22% Candidate shortlists. Increase in recruiter shortlists in profile-optimization experiments.
−20% Candidate rejections. Reduction observed in the profile-optimization experiments.
Better AI answers. Lower model cost.
A production model was retiring, and its proposed replacement threatened a threefold cost increase. I compared three providers, checked answers against the candidate’s actual experience, revised prompts, and shipped the best-performing integration.
Compare answers.
Check the evidence.
Built inventory & invoicing APIs with Node.js.
The opening asks for Kubernetes.Relevant work and the original technology are retained.
Revise the prompt. Test again.
Answers are illustrative, not employer outputs. Cross-model judging supports the comparison; source checks still determine whether a claim is justified.
How I built itRead the engineering details
The evaluations exposed a tension between preventing invention and preserving useful source material. A stricter instruction could remove content it should have kept. Cross-model judging provided another perspective, but I still checked the evidence and used deterministic validation for rules with a definite answer.
- LLM evaluation & test design
- Turn a vague quality question into repeatable comparisons.
- Cross-model judging & validation
- Combine evidence-based review with checks for clear rules.
- Prompt engineering
- Use observed failure patterns to improve the next version.
+43% Evaluation quality. Quality improvement from the selected model integration after comparing three LLM providers.
−23% Model cost. Lower cost for the selected integration in the model comparison.
Recruiters could see why a candidate matched.
A match score didn’t explain which experience supported it. I connected job requirements to supporting career experience and made the reasons inspectable. Missing evidence stayed distinct from a mismatch.
Job requirement.
Supporting experience.
Platform
engineer
Build reliable services.
Work across product and data.
Alex
Okafor
Software engineer
Backend development
Built inventory APIs and background jobs to synchronize product records.
Relevant work supports the requirement. Each factor preserves the distinction between a match, a mismatch, and an unanswered question.
How I built itRead the engineering details
A simpler display would have been misleading if it flattened every factor into yes or no. The work preserved match, mismatch, and unknown states, then let people inspect the evidence without making them read everything at once.
- Vue, TypeScript & shared components
- Bring explanations into product views without duplicating their behavior.
- API contracts & data consistency
- Keep the meaning of candidate information aligned across services and screens.
- Evidence attribution
- Connect a skill or claim to the experience that supports it.
Engineers could trace failed AI calls.
Failed model calls and poor search results were difficult to investigate. I recorded model attempts and failures, clarified product metrics, and sent the search context and displayed candidates to the team’s review workflow.
The successful result.
The failed attempt.
request / sample-1042Result returnedBoth attempts stay visible.
A successful fallback does not erase a failure.
The failed attempt
The provider failure remains in the trace, even when a later attempt succeeds.
Model tracing exposes failed calls. Search context is sent separately to the team’s review workflow, without blocking the recruiter’s search.
How I built itRead the engineering details
The review team needed the result the user actually saw. I worked on carrying that context into the review handoff while keeping a delivery failure from interrupting search. For metrics, I checked what each count represented and kept incomplete periods from distorting comparisons.
- OpenTelemetry & Langfuse
- Record the parts of model operations, their timing, and failures.
- SQL & event measurement
- Count a defined population and compare matching time periods.
- n8n integration & HTTP
- Send useful search context into the review team’s workflow without owning or recreating their receiver.
Interrupted applications could continue without losing progress.
Search, résumé imports, uploads, and messaging could fail partway through an application. I traced the failures across the browser and backend, then repaired recovery and data handling while preserving completed work.
Retry the failed step.
Keep completed work.
Alex Okafor
Backend APIs.
Background processing.
Search interrupted
The search paused. Imported information and the saved profile remain available.
This example illustrates recovery that preserves progress. Import, upload, messaging, and service-resource failures each needed a correction at the place that owned the problem.
How I built itRead the engineering details
A retry is useful when a dependency fails temporarily; it cannot repair an incorrect data merge or a client created too often. The investigation mattered because each symptom needed a different correction, with existing user progress preserved where possible.
- Debugging & resource lifecycle
- Find the owner of a failure and correct it without introducing unnecessary machinery.
- Asynchronous recovery
- Handle interrupted work while preserving completed progress.
- Browser APIs & integration
- Keep uploads, recordings, and messaging compatible with the surrounding product.
FROM SOMEONE I WORKED WITH“Not just using AI but keeping a critical mindset in order to get the best out of it.”
Gus GarcíaProduct Engineer at Torre.ai · Direct manager