Public employer and investor demo

InterviewCoach

A working Android and backend product showing how agentic engineering can move from product UX to release gates, observability, scale testing, and model-cost governance.

Run 10-minute demo Open runtime status
Live proof stack Android app, FastAPI/LangGraph backend, Postgres/Qdrant, LangSmith, Prometheus, release scripts, and public-safe ops pages.
Agentic SDLC OpenClaw/Codex-driven implementation with human review gates
Mobile release Signed Android artifact, screenshots, metadata, and Play gates
Scale path Deterministic load proof plus switchable live-model runtime
Cost control Zero-spend stress tests and capped live-model sampling

What this proves.

InterviewCoach is the public narrative layer for a larger build discipline: product empathy, AI architecture, mobile release quality, and operational truth in one maintainable system.

Product

Useful mobile practice

The app helps engineers prepare for production AI interviews through chat, scoring, concrete gaps, and next practice missions.

Platform

Agentic delivery loop

The repo carries docs, gates, memory, release packets, trace smokes, and public proof surfaces so the build can continue without losing context.

Business

Scalable app path

The current Android-first proof can expand toward broader mobile coaching while keeping runtime choices, costs, privacy, and support surfaces explicit.

Maintainable system blueprint.

The main goal is not a one-off demo. It is a scalable mobile product core where product, AI, release, and operations stay connected.

01Android client

Chat-first preparation, results review, release screenshots, privacy/support links, and Play-ready metadata.

02AI backend

FastAPI orchestration, LangGraph flow boundaries, retrieval context, report generation, and provider switching.

03Data layer

Postgres records, Qdrant retrieval path, local deterministic mode, and private candidate/job data kept out of public pages.

04Operational proof

LangSmith traces, Prometheus-style metrics, runtime summaries, release gates, checksums, and launch-readiness scripts.

Five questions, direct answers.

Use this page as a short technical walkthrough before opening the deeper repo docs or protected investor materials.

  1. 01
    Agentic architecture

    The system combines a product backend, Android client, observability surfaces, and OpenClaw/Codex as the implementation agent. State is persisted in backend records, task context is carried through repo docs and memory dossiers, and hand-offs are controlled through review gates instead of unbounded autonomous writes.

  2. 02
    RAG for mobile codebases

    A Swift/Kotlin meta-repo should be indexed by module, architectural layer, test surface, symbol ownership, and dependency graph. Chunks should preserve class/function boundaries and carry metadata for MVVM roles, SOLID concerns, platform, package, Gradle/Xcode target, and related tests.

  3. 03
    CI/CD activation

    An agentic pre-flight can run in GitHub Actions before merge: inspect diff, retrieve nearby patterns, generate JUnit/XCTest candidates, flag architecture violations, and publish PR comments. Risky changes stay advisory until a human accepts the generated patch.

  4. 04
    Observability and metrics

    AI tool events should be normalized into delivery events and linked to Jira issue keys, PRs, trace ids, model cost, and review outcomes. DORA movement is measured by comparing similar work classes over time, with manual baselines kept separate from AI-assisted flows.

  5. 05
    Financial governance

    A model-router middleware selects deterministic runtime, Bedrock, Gemini, or OpenAI by task complexity, latency target, privacy boundary, token budget, and required reasoning quality. Stress tests avoid paid models; live sampling is capped and auditable.

Ten-minute demo path.

Show product, runtime, delivery proof, and honest boundaries without exposing secrets or raw user data.

StartOpen the Android APK, show setup/chat/review screenshots, then explain the recruiter-style interview loop.
RuntimeOpen ops runtime and metrics summaries to show live backend status, model path, and public-safe counters.
DeliveryShow release notes, Play metadata gates, signed artifacts, checksum, and the current Play upload blocker.
ScaleExplain deterministic load testing first, then tiny capped live-model samples for trace proof and cost governance.
CollaborationExternal Android Studio can work on feature branches while OpenClaw/Codex owns repo checks, docs, release gates, and integration review.

Why it fits employers and investors.

The same proof works for hiring and diligence because it connects implementation decisions to product value, release confidence, and future scale.

EmployersEvidence of hands-on AI engineering: product UX, RAG strategy, mobile architecture, tests, release scripts, and observability.
TeamsA pattern for agentic pre-flight review, repo-aware code assistance, traceable automation, and human-controlled integration.
InvestorsA practical mobile app thesis: help people improve interview performance while proving cost, privacy, operations, and rollout discipline.
Next scaleBroaden coaching scenarios, add iOS/XCUITest, deepen codebase RAG, activate paid-model scale only behind explicit budgets.

Public-safe proof surfaces.

These links are safe to share in an employer conversation. Protected traces and private CV/job data remain outside public pages.

Product siteRelease artifact, screenshots, privacy, support, and release notes. Test evidenceJob-match runs, interview situations, scores, and source job URLs. Investor roomTechnical diligence overview and scale/cost evidence. Ops summaryPublic-safe backend and dependency status. RuntimeAgent runtime, model mode, tracing status, and surface links.

Current boundaries.

The demo is strongest when it is precise about what is already proven and what is intentionally gated.

Proven now

Android release artifact, public site, backend orchestration, deterministic load path, LangSmith trace smoke, investor/employer answer material, and repo-level gates.

Blocked by access

Actual Google Play draft upload waits for the Android Publisher service-account JSON or authenticated Play Console access.

Gated by spend

AWS Bedrock execution and larger live-model scale tests should start only after region, model access, and a small budget alert are confirmed.

Roadmap, not claim

iOS/XCUITest and full Swift meta-repo indexing are described as the next implementation pattern, not as already shipped proof.