Case file 01 / Independent Project

OptiTalent

Evidence-first recruiting workflows

A multi-tenant recruitment product that turns candidate search, evidence review, shortlist decisions, controlled sharing, and recruiter feedback into one coherent workflow.

Classification: Independent Project
Delivery status: Independent product / active validation
Case: 01 / 07

GitHub Actions / CI

Four runs, with passed and skipped jobs visible.

GitHub main run 36028002118: impact, security, contracts-and-release and required passed; browser skipped by impact routing — 01
**Post-#410 main gate**
Run 36028002118. Redrawn from a private GitHub receipt; browser was skipped.

GitHub run 36012881237: impact, security, contracts, browser and required jobs passed — 02
**Required gate**
Run 36012881237. All five shown jobs passed on that commit.

GitHub run 35997399618: Rust contracts, hosted runtime tests and release builders passed — 03
**Rust and release build**
Run 35997399618. This checks source and artifacts, not live deployment.

GitHub run 36016601619: browser regressions and synthetic evidence passed — 04
**Browser regression**
Run 36016601619. Synthetic browser evidence, with no private candidate records.

Runner / release

Checks and deployment have separate boundaries.

Annotated runner path: GitHub job label, separate browser and release ARC listeners and scale sets, pod templates, k3s scheduling and result reporting — 01
**Self-hosted runner**
PR #410 defines the shared heavy slot. The diagram marks live Helm and scheduling acceptance as unverified.

Annotated 12-step release path: GitHub impact routing and required CI checks, then Gitea exact-SHA mirror, immutable build, staging, approval, manual promotion and readback — 02
**Release path**
Steps 01–06 are GitHub source CI; 07–12 are separate Gitea release control. Notes explain skip and approval rules.

Hosted MCP

Sensitive actions need a separate browser approval.

The problem

Recruiting teams need to move quickly without turning matching into an opaque score or losing the evidence behind a decision.

System arrangement

How the system works

Confirmed role constraints fan into exact and semantic recall lanes. Candidate-level RRF builds the pool, ColBERT refines it, and evidence policy controls what can reach a recruiter shortlist.

01
SourceRole intent + constraints
Confirmed eligibility compiles first
02
CapabilityPGlite exact + lexical
Canonical facts, filters, and FTS
03
CapabilityQdrant dense + sparse
BGE-M3 versioned derived index
04
CapabilityCandidate-level RRF
Fuses only executed recall lanes
05
CapabilityColBERT MaxSim
Reranks the fused candidate pool
06
ControlEvidence policy
Unknown stays unknown, not excluded
07
OutcomeRecruiter shortlist
Evidence remains reviewable
08
ControlTenant-local compute
Redacted subjects stay close to data

My role

I defined the product boundaries, tenant model, evidence contracts, recruiter workflow, review gates, and release acceptance. AI agents produced most implementation code under those constraints.

Capabilities

01
Hybrid lexical and dense candidate retrieval
02
Evidence-preserving reciprocal-rank fusion
03
Local embedding and Qdrant recall runtime
04
Candidate evidence and explainable match signals
05
Draft and approved shortlist states
06
Controlled sharing and feedback
07
CLI, MCP, desktop, and web access paths

Product decisions

01
Prefer evidence-first retrieval and explicit unknowns over a single opaque LLM judgment.
02
Place bounded embedding and retrieval work close to tenant data while preserving server-side policy and review authority.
03
Keep the policy and lifecycle authority behind stable contracts while interfaces render approved state.
04
Treat agent and human entry points as adapters over the same capabilities.

Evidence

•
Public product surface at app.optitalent.cc.
•
Implemented candidate search, candidate evidence, shortlist, and sharing contracts.
•
Tenant-scoped retrieval and explicit review states are present in the product runtime.
•
LightGBM LambdaRank was evaluated against deterministic baselines on a frozen retrieval protocol and was not selected as the production winner for that feature shape.

My part

I set the product, evidence, approval, tenant, and release boundaries. Coding agents wrote much of the implementation; I reviewed the behavior and delivered system. This is an independent product, not a team-management claim.

What is verified

The GitHub extracts above show exact passed or skipped jobs, not every test on every commit. A September 2026 production readback found healthy gateway and backend containers, MCP enabled, and anonymous MCP requests denied with 401. It did not exercise an authenticated end-to-end user journey.

OptiTalent has no users yet. I claim no adoption, revenue, hiring outcome, or model-quality improvement.