Independent backend builder

Homeowner Permit Compliance Checker

Built Python ingestion with Apify, a shared PostgreSQL rule schema, FastAPI project-attribute lookups, and an OpenAI conversational agent that calls database tools and explains retrieved requirements.

Open PreClear(Canonical app: Preclear/Preclear-Astra. Live data and model configuration are pending.)Browse canonical source

Role: Independent backend builder
Year: Jan–Mar 2026 · Updated Oct 2026
Scope: Python, PostgreSQL, FastAPI, JavaScript, Apify, OpenAI
Evidence: Implementation added October 2026; local tests, cloud validation pending

Problem + constraints

Homeowners need to connect project details with changing jurisdiction-specific requirements and understand the sources behind the answer.

Municipal raw documents and chunks are stored in Supabase. Its database-size quota currently blocks verification of the résumé’s 50+ source-site count.
Canonical implementation: Preclear/Preclear-Astra. API and UI preview checks passed; live database and model configuration remain pending.
Initial project: January–March 2026. Further development in October 2026 added FastAPI lookups, Apify ingestion and the conversational agent.

Contribution boundary

My contribution

Added Python and Apify collection with source evidence, normalized rule candidates and reviewed PostgreSQL ingestion.
Built FastAPI endpoints that map project attributes to jurisdiction-specific permit rules.
Built a conversational OpenAI agent that calls a validated database lookup tool and explains retrieved requirements in plain language.

Outside my claim

Production adoption, uptime, accuracy or latency guarantees.
Retrospective proof that newly added features existed during the project dates.

System design, explained

The 30-second explanation

Think of this as a research assistant with a checked rulebook. First, municipal documents are collected and a person reviews the extracted rules. Then a homeowner asks a question. The assistant looks up the rules for the selected location and explains the result with sources.

Implementation statusCanonical app: Preclear/Preclear-Astra. UI and API code are deployed; auth, database and model configuration still prevent a verified live answer. The 50+ source-site figure is not verified.

Follow the data

A map of the moving parts.

Select any stage to inspect its input, work, output and failure behavior.

Example, not a live resultIllustrative question: “Do I need a permit for a 120-square-foot shed?” The homeowner selects a jurisdiction; no permit determination below is a real municipal answer.

01 / Prepare the rulebook

Ingestion is separate from the homeowner’s request. Existing Supabase documents can be reused.

Alternative source inputs

Reviewed rules → PostgreSQL → available to the lookup in lane 02

02 / Answer a question

The model translates the question; application code evaluates the rules.

Tool result → back to OpenAI for explanation → response routes in lane 03

03 / Return an honest result

The model receives the tool result and explains it; it cannot approve a permit.

Lookup outcome · two possible routes

Plain-language response and retrieved source links → homeowner interface

Technology, with a purpose

What each tool is responsible for.

Languages, frameworks and services play different roles. Here is why each belongs in this system.

Technology roles and design rationale
Technology	Responsibility	Why this choice
React + TypeScriptBrowser interface	Question entry, jurisdiction selection, conversation and source display.	Components organize the interface; types describe data exchanged with the API.
ViteFrontend build tool	Develops and bundles the React/JavaScript application.	Produces the static frontend served with the API deployment.
Python + FastAPIApplication service	Validates requests, verifies sessions and exposes chat and deterministic lookup endpoints.	Keeps business rules and ingestion in Python with explicit API contracts.
ApifyCollection service	Collects specified municipal pages for ingestion.	Separates page acquisition from rule extraction and review.
OpenAILanguage model API	Drafts ingestion candidates and interprets homeowner questions using read-only tools.	Handles varied wording while decisions remain in the rule evaluator.
PostgreSQL / SupabaseData + authentication	Stores raw documents, chunks and reviewed rules; verifies homeowner sessions.	A relational schema preserves jurisdiction, conditions and source provenance.
VercelHosting	Serves the React app and same-origin FastAPI Python function.	Keeps browser and API under one origin; hosting alone does not configure model or data access.

Interview walkthrough

How I explain the design.

Start with the problem, follow one input, then explain the tradeoff.

01
I separate building the rulebook from answering a homeowner. Crawled or existing documents become candidates, then a human reviews them.
02
For a question, the UI sends a jurisdiction and authenticated request. OpenAI turns everyday wording into a constrained lookup tool call.
03
Python evaluates reviewed PostgreSQL rules. The model explains that result and cites its sources; it cannot change the jurisdiction or write to the database.
04
The most important design choice is how we handle uncertainty: missing details or conflicting rules mean municipal verification, not a guessed answer.

Decisions + tradeoffs

D-01

Preserve conditions and source provenance

Normalized rules retain compound conditions, citations, source hashes, review dates and expiry.

D-02

Keep decisions in the lookup service

FastAPI evaluates all applicable reviewed rules; missing inputs and conflicts require jurisdiction verification.

D-03

Constrain agent tools

The agent can call one read-only lookup tool, with jurisdiction fixed by the server and bounded tool rounds.

Evidence ledger

Claims are labeled by what a reader can inspect now. Withheld items include the measurement needed before publication.

E-01

Implementation and local verification

Limited

Canonical React/Vite and FastAPI implementation added in October 2026: 24 Python tests, production build, and deployed API authentication checks passed. Live data/provider validation remains pending.

Method: Local domain, orchestration and API checks; provider responses are mocked.
Source / path: Repository implementation and test files

E-02

Deployment and measured outcomes

Withheld

Vercel preview serves the API and interface. No successful live model/database integration, production benchmark or coverage measurement is claimed.

Method: Verify deployed services and publish dated measurement artifacts before adding outcome claims.
Source / path: Explicit verification boundary

Not production-ready

Current limits

Municipal raw documents and chunks are stored in Supabase. Its database-size quota currently blocks verification of the résumé’s 50+ source-site count.
Canonical implementation: Preclear/Preclear-Astra. API and UI preview checks passed; live database and model configuration remain pending.
Initial project: January–March 2026. Further development in October 2026 added FastAPI lookups, Apify ingestion and the conversational agent.

Redesign for scale

Run deployed integration checks with representative inputs and record dated results.
Monitor failures, retries and operating costs with a defined ownership process.