Python Engineer — Production Hardening · Pricing Engine · WebSockets · AI Review Layer

Hi. This project is in my zone — automation, integrations, backend logic, and making the workflow reliable end-to-end.
I can take ownership of the system, not just patch one part of it.
Relevant experience:
- Built lead generation and API-driven automation pipelines focused on filtering, enrichment, and reliable business workflows.
If the scope is clear, I can move fast and build it cleanly so it doesn't need babysitting after launch.
If you want, I can also outline the implementation approach before we start.

Hello! My name is Alex, and I represent the NC-1 development group. For over five years, we have been building websites, mobile applications, ERP/CRM systems, and other e-commerce products. I would like to offer the outstaff services of our Full Stack, Middle+ engineer. He has over 3 years of extensive experience in IT, with a particular focus on web development, e-commerce, and solution/product creation.

Based on the technical brief provided, here is our detailed proposal for your Intelligent Vehicle Platform:

1. Total Estimated Hours
For the complete production hardening, removal of placeholders, and implementation of the required deterministic logic, we estimate a total of 120–155 hours.

2. Breakdown by Workstream
WebSocket System (20–24 hours): Implementing broadcast_to_client, transitioning to an async Redis client (pub/sub or streams), maintaining the user-to-socket mapping, and ensuring multi-instance stability.
…
Pricing Engine & Similar Listings (30–38 hours): Replacing random sampling with a weighted scoring model, implementing tiered comparable selection rules, outlier filtering, and strict tie-breaking logic.

Smart Review AI & Cache Layer (22–26 hours): Replacing mocks with real vehicle fingerprint generation, web evidence retrieval, LLM synthesis, and a versioned cache layer with TTL invalidation.

DB, Backend Stabilization & Search (20–24 hours): Eliminating crash paths, fixing SQL predicates, enforcing ownership/uniqueness constraints, and replacing placeholder search with validated filter-based search.

Integrations & Analytics (12–16 hours): Reconnecting image intelligence signals into the DAG, resolving Facebook module proxy issues, and ensuring schema alignment across parsers.

Testing & CI (16–20 hours): Implementing full pytest-asyncio coverage for all critical flows and ensuring the suite is fully CI-compatible.

3. Risks and Unknowns
Image vs. Text Conflicts: Resolving discrepancies when image classifier data contradicts text metadata requires a predefined hierarchy of truth.

Facebook Integration: Existing account and proxy issues may require additional infrastructure troubleshooting beyond standard code fixes.

External API Latency: Real-time web evidence retrieval for AI reviews can introduce latency; we recommend implementing optimized fallback or background processing states.

4. Proposed Milestone Plan
Phase 1: Stabilization & DB Integrity: Removing crash paths, fixing SQL bugs, and adding the necessary DB indexes and constr

Здравствуйте, Max.

Прочитал полный бриф — он нетипично чётко структурирован для проекта по укреплению системы. Архитектура готова, не хватает детерминизма, надёжности и реальных реализаций там, где сейчас заглушки. Ниже ответ по Вашему 6-пунктовому формату.

1. Общая оценка: 100–130 часов.
Нижняя граница — потому что границы скоупа в брифе чистые. Верхняя — если заглушек больше чем видно снаружи или у listing_margin есть data-quality долг, который не виден из брифа.

2. Разбивка по workstreams:
— Стабилизация backend + укрепление БД (null safety, индексы, constraints, SQL predicate bugs, structured logging): 12–15 ч
… — WebSocket reliability (broadcast_to_client, маппинг user_id ↔ active sockets, async Redis pub/sub, multi-instance, pytest-asyncio): 18–22 ч
— Детерминированный pricing engine listing_margin (tiered comparable selection, widening fallback, outlier filtering, strict rounding, persisted benchmark metadata, backfill script): 15–18 ч
— Детерминированный similarity engine (weighted scoring, configurable weights, hard filters + widening fallback, tie-breaking, indexed queries): 12–15 ч
— Analytics graph — image intelligence integration (reconnect classifier в DAG, разрешение конфликта image-vs-text, safe fallback, observability): 8–10 ч
— Smart Review AI layer (vehicle fingerprint generation, web evidence retrieval, evidence-only LLM synthesis, fingerprint-keyed cache, TTL + version invalidation, forced refresh): 18–22 ч
— Notification lifecycle (update predicate fix, per-user ownership, orphan cleanup, transactional consistency, DB uniqueness): 6–8 ч
— Catalog search replacement (real filter-based search, validated pagination, integration tests): 4–6 ч
— Facebook integration (reconnect модуля, accounts/proxy, schema alignment): 5–7 ч
— CI + полное async-тестовое покрытие: 10–12 ч

3. Риски и неопределённости:
— Иерархия разрешения image-vs-text конфликта (workstream 5) — если классификатор говорит "SUV", а text matcher говорит "sedan", кто выигрывает? Без явных бизнес-правил детерминированный engine становится недетерминированным на границах. Правила нужно зафиксировать до того как я трону этот модуль.
— Web evidence retrieval для Smart Review — сырой скрапинг хрупок под Cloudflare / rate limits. Ранняя развилка: платный SERP API (SerpApi / Tavily / Brave) vs свои парсеры. Влияет и на надёжность, и на ongoing cost.
— Facebook модуль — "accounts/proxy issues" может скрывать token invalidation или CAPTCHA loops. Нужен scoped spike до фиксирования бюджета на workstream 9.
— Объём backfill — как только pricing станет детерминированным, исторические строки скорее всего потребуют пересчёта. Количество строк из брифа не видно.
— LLM cost ceiling — fingerprint cache снижает, но forced-refresh paths нужны budget guards.

4. План этапов:
— M0 — Короткий аудит (4–6 ч, фиксированная цена): клонирую репо, размечаю текущие placeholders, подтверждаю оценки workstreams против реального кода, флагаю скрытую связанность. Нет обязательства на полный engagement если риски окажутся больше. Низкорисковый первый шаг для обеих сторон.
— M1 — Фундамент: workstream 1 (стабилизация) + 7 (notifications) + 8 (catalog search). Чистая база до engines.
— M2 — Детерминированные engines: workstream 3 (pricing) + 4 (similarity) + backfill scripts. Checkpoint: одинаковый вход → одинаковый выход, assert-ится в тестах.
— M3 — Realtime + AI: workstream 2 (WebSocket hardening) + 6 (Smart Review с cache). Интеграционные тесты против реального Redis.
— M4 — Integrations + CI: workstream 5 (image DAG) + 9 (Facebook) + полный тест-сьют + CI + migration notes.

Каждый milestone доставляет рабочий инкремент с проходящими тестами, не половинчатое состояние.

5. Стратегия тестирования:
— pytest + pytest-asyncio, CI-compatible с M1.
— Детерминированные unit-тесты для pricing и similarity — одинаковый вход должен всегда давать одинаковый выход, assert-ится напрямую.
— Integration тесты против реального PostgreSQL (не sqlite), через testcontainers или docker-compose fixture.
— Async WebSocket тесты с реальным Redis для проверки multi-instance pub/sub — unit mocks тут недостаточно.
— Contract тесты для Smart Review: структура и evidence-binding, не содержание LLM output.
— Migration тесты: apply + rollback + re-apply чисто.

6. Требуемые доступы:
— Репо (read/write на feature branch)
— Текущая схема + история миграций Alembic
— Staging PostgreSQL + Redis или воспроизводимый docker-compose
— LLM provider keys + провайдер + cost ceiling per request
— Текущие credentials Facebook модуля или тестовые аккаунты
— CI config (GitHub Actions / GitLab / другой) с правами на модификацию
— Примеры failing cases для pricing / similarity / AI review если уже трекаются

Стек ежедневно: FastAPI · async SQLAlchemy · async Redis · Postgres · pytest-asyncio · Docker. Это ровно та работа которой я занимаюсь — укрепление production-систем, не greenfield.

Готов начать с M0 аудит (fixed, 4–6 часов) как низкорисковый первый шаг. Получите мои подтверждённые оценки против реального кода до коммита на полные milestones.

Hi,

I’ve reviewed your technical brief carefully — the system is well-structured, and the scope is very clear: this is a classic transition from a feature-complete prototype to a deterministic, production-grade platform. This is exactly the kind of work I specialize in.

1. Estimated Total Effort

~140–180 hours total

(This assumes no major architectural rewrites and that current modules are reasonably isolated.)
…
2. Risks / Unknowns
Current DB state (data quality, duplicates, missing relations)
How “deterministic” current matching logic actually is (may require deeper refactor)
External dependencies for web evidence retrieval (rate limits, scraping reliability)
Existing Redis/WebSocket implementation quality (possible rewrite vs patch)
Facebook module stability (auth/proxy issues can be time-consuming)
LLM cost/performance constraints depending on traffic

3. Proposed Milestones

Milestone 1: Stabilization Layer

DB fixes, constraints, migrations
Remove placeholders
Logging + crash safety

Milestone 2: Real-time System

WebSockets redesign (auth + Redis)
Notification lifecycle completion

Milestone 3: Deterministic Engines

Pricing engine
Similarity engine
Backfill scripts

Milestone 4: AI Smart Review

Evidence retrieval + LLM pipeline
Cache layer + invalidation

Milestone 5: Search & Integrations

Catalog search
Facebook integration

Milestone 6: Testing & CI

Full async test coverage
CI pipeline stabilization
5. Testing Approach
Unit tests for deterministic logic (pricing, scoring)
Integration tests for DB + API flows
Async tests for WebSockets and Redis (pytest-asyncio)
Snapshot-style tests for AI responses (structure, not content)
Backfill validation scripts to ensure consistency
CI-ready test suite with reproducible environments
6. Required Access
Codebase (repo + branches)
Database access (read + staging write)
Redis instance (or staging equivalent)
API keys for LLM provider
Access to current WebSocket infra
Facebook module credentials/config
CI environment (or ability to configure one)
Relevant Experience
Production hardening of FastAPI systems
Designing deterministic ranking/scoring systems
Async architectures (Redis, WebSockets, event-driven flows)
LLM pipelines with caching and cost control
Refactoring prototype systems into stable production services

If helpful, I can start with a short audit phase (4–6 hours) to validate estimates and identify any hidden complexity before committing to full milestones.

Best regards,
Oleh

I worked on Poseidon (https://poseidon.codezerogroup.com) — enterprise backend Python/FastAPI with data pipelines for CodeZero Group.

I read the full brief. 9 streams, each with separate dependencies and risks.

What I will do:

Stream 1 — Backend Stabilization (3d): null handling, DB integrity, structured logging, SQL bugs, ownership constraints.
Stream 2 — WebSocket + Auth (4d): broadcast mapping, async Redis pub/sub multi-instance, pytest-asyncio CI-compatible.
Stream 3 — Pricing Engine (4d): deterministic fallback, outlier filtering, rounding policy, benchmark metadata, unit tests.
… Stream 4 — Similar Listings (3d): weighted scoring, hard filters + widening fallback, indexed queries, tie-breaking.
Stream 5 — Image Intelligence (3d): reconnect classifier in DAG, merge image signals, safe fallback + observability.
Stream 6 — Smart Review AI (5d): vehicle fingerprint, LLM evidence-only synthesis, cache TTL + version invalidation.
Stream 7 — Notification Lifecycle (2d): update predicate, per-user ownership, orphaned cleanup, endpoint tests.
Stream 8 — Catalog Search (2d): filter-based search, validated pagination, integration tests.
Stream 9 — Facebook Integration (2d): module integration, accounts/proxy resolution, schema alignment.

Total 28d + 3d CI = 31 days. Tests: pytest + pytest-asyncio, TestClient + PostgreSQL + Redis mock.

--- OPTIONS ---

- Option A (5 streams): 3500 USD (31 days) — Streams 1+2+3+4+6: stabilization + WebSocket + engines + AI review
- Option B (Full system): 5600 USD (42 days) — all 9 streams + CI + backfill + documentation — best scope/price ratio
- Option C (System + architecture): 7280 USD (56 days) — everything from B + code review + architecture documentation + 30d support

Execution time: 31 days. I need: repo access, .env, listing data, LLM brief, FB App Review status.

Portfolio:
- https://poseidon.codezerogroup.com — enterprise Python, FastAPI backend, data pipeline
- https://ou-uv.com — Flask/Python CMS, API integrations, multilingualism
- https://codezerogroup.com — B2B, multi-module web systems, backend

8 years in Python / AI — from scripts to agent systems with enterprise integrations.

Write to me, I will send a detailed plan stream by stream.

Since I am new to the freelancehunt service and want to quickly gain a few initial projects for my portfolio, I offer a 15% discount for the first 5 clients. The offer is valid until 5 orders are obtained.

Show original

Translate

Hello! I have experience with FastAPI, SQLAlchemy, Redis, WebSockets, PostgreSQL, and LLM. I will conduct an audit, fix SQL errors and indexes, create deterministic pricing and similarity algorithms, set up WebSockets with Redis pub/sub and authorization, replace the AI review mockup with a real one (web proofs → LLM → cache with TTL), add Facebook integration, and cover everything with pytest-asyncio in CI. I work in stages. Details in private!

Show original

Translate

Hello! I am an individual developer with 4 years of experience in building complex Backend systems on FastAPI and PostgreSQL, so I specialize in transitioning from MVP fillers to production-ready architecture. My approach is based on replacing "random" logic with strict mathematical models (weighted evaluation, outlier filtering using Tukey's method or Z-score) and ensuring horizontal scalability through Redis Pub/Sub for WebSocket connections. I will fix the database structure, implement deterministic ranking algorithms, and realize integration with AI through cached synthesis with version control, ensuring 100% coverage of critical asynchronous streams with tests in CI/CD. I work through a sole proprietorship, focusing on SQL performance and authorization security; I am ready to study the technical brief and propose architectural solutions for each workflow. My works: https://3magency.co/, https://jk-solution.com.ua/, https://farfieworldwide.com/, Behance.

Show original

Translate

Dear Max Scat,

Thank you for sharing the brief. This looks like a system that already works but needs to be made stable, predictable, and ready for production — which is exactly the kind of work I do.

In my recent projects, I’ve taken similar backends and removed random or placeholder logic, replacing them with clear and deterministic behavior, especially in scoring and ranking systems. I’ve also built WebSocket systems with Redis that handle per-user connections reliably across multiple instances, and turned mock AI features into real pipelines using data retrieval, controlled LLM output, and caching with proper invalidation.

For your platform, I already have a clear plan for how to handle the key parts — making pricing and similarity fully deterministic, improving WebSocket reliability, and building a smart review system that uses real data with caching. I’m also used to cleaning up FastAPI and SQLAlchemy code by fixing queries, adding constraints and indexes, and making sure everything is well-tested, including async flows in CI.

Based on the scope, I estimate the work can be completed in 1~2 weeks.
…
This project fits my experience very well, and I’m confident I can help make the system stable and production-ready.

Happy to share a detailed plan if you’d like to continue.

Best regards,
Jeo

ready to help you out

will share prevoius work in chat to make sure we are match



--------------------------------------------------------

Hi.

I'm a senior Python developer with 10+ years in production projects. Most of my career I've worked with existing codebases — stepping into other people's systems, figuring out how they're wired, and getting them to a state worth being proud of. One example: I built and ran a taxi dispatch platform solo for 6 years — 130k+ orders/month, 900+ drivers online simultaneously, real-time GPS sync every second. That's the kind of production pressure I'm used to.

I've also worked with AI integrations: LLM pipelines, web retrieval, cache layers with versioning — exactly what you're describing in the Smart Review section.

I read through the job post and the brief. Giving exact numbers without an initial code review is always a bit of a guess: the brief clearly describes *what* needs to be done, but doesn't tell me how deep the stubs go, what shape the migration debt is in, or what's really going on with the Facebook module and image pipeline. Any one of those can shift the estimate significantly.

That said, based on the described scope, my working estimate is: 180–240 hours
…
Breakdown by module:

1. WebSockets / Redis — 35–45 h. broadcast_to_client, user↔socket mapping, async pub/sub, multi-instance safety, async tests.
2. Pricing engine — 30–35 h. Tiered selection, outlier filtering, deterministic fallback, backfill script.
3. Similarity engine — 25–30 h. Weighted scoring, configurable weights, tie-breaking, indexed queries.
4. AI Smart Review — 30–40 h. Fingerprint → web retrieval → LLM → cache with TTL and versioning.
5. DB / SQL hardening — 25–30 h. Indexes, constraints, ownership enforcement, null safety, clean migrations.
6. Tests / CI — 20–25 h. pytest-asyncio, coverage of critical flows, CI-compatible suite.

Facebook and image pipeline I'll estimate separately — once I can see the state of those modules.

To get started I'll need:
— repo access
— current Alembic migrations and DB schema
— .env.example or a list of env variables
— infra overview (Docker, staging, number of instances)
— LLM provider and model for Smart Review
— current CI config

I have experience working with production backend on FastAPI, PostgreSQL, Redis, and WebSockets. I have been involved in stabilizing existing systems, removing unstable logic, building deterministic algorithms, setting up asynchronous tests, and integrating with AI APIs.

I have reviewed the brief, understand the scope and tasks. I can complete the stabilization of the backend, bring the WebSocket part to production level, implement deterministic logic for pricing and similarity, replace mock AI with a real implementation with caching, and also tidy up the database and test coverage.

I initially estimate the scope to be within 90–140 hours, but I will provide a more accurate estimate after reviewing the code.

I would start with a brief audit and setting up the environment, after which I would sequentially close the main blocks and bring the system to a stable state.

To start, I need access to the repository, environment, database, and CI, as well as a brief technical overview of the current architecture.
…
I am ready to discuss the details and quickly get started on the work.

Show original

Translate

Вітаю!

Маю досвід роботи з FastAPI, SQLAlchemy, PostgreSQL, Redis, WebSockets, AI/API інтеграціями та production backend-системами*, включно з рефакторингом існуючих рішень.
Використовуваний стек (рекомендований):

Backend:Python, FastAPI, SQLAlchemy
DB / Cache: PostgreSQL, Redis (async)
Integrations:** WebSockets, LLM API, Facebook integration
Testing / Infra Pytest, pytest-asyncio, CI/CD, Docker
…
**Ризики / unknowns, які бачу вже зараз:

* поточний стан legacy-коду і рівень зв’язаності модулів
* наскільки глибоко placeholder-логіка прошита в бізнес-флоу
* стан міграцій та реальні проблеми з даними в PostgreSQL
* реальний формат “web evidence fetch” для AI review
* поточна схема авторизації у WebSocket шарі
* обсяг edge-cases у pricing/similarity логіці

**Пропонована milestone-структура:**

1. **Audit + stabilization plan**
короткий технічний аудит, фіксація ризиків, уточнення acceptance criteria
2. **Core logic hardening**
deterministic pricing + similarity ranking
3. **Realtime layer**
WebSockets + Redis pub/sub + auth isolation
4. **AI review module**
evidence fetch + synthesis + caching/versioning
5. **DB hardening**
constraints, indexes, migrations, cleanup
6. **Testing + CI + final integration**
async tests, regression checks, Facebook module, release prep

**Підхід до тестування:**

* unit-тести на pricing / similarity rules
* integration-тести на DB + migrations
* async тести на WebSockets/pub-sub
* contract-тести на AI review pipeline
* smoke/regression suite в CI
* окремо перевірка deterministic output для критичних сценаріїв

Що потрібно від вас для старту:

доступ до репозиторію
технічний бриф / scope doc
доступ до staging/dev environment
.env.example або список потрібних сервісів
поточна схема БД / міграції
приклади проблемних кейсів по pricing / similarity / AI review
доступи або sandbox по Facebook integration та LLM provider

Готовий підключитися поетапно і закрити це як production hardening.

З повагою,
Андрій

Hi Max,

I reviewed the brief carefully. This is the same production-hardening scope I would treat as an existing system stabilization task, not a rebuild. My estimate stays in the same range as before.

1) Estimated total hours
- 150–165 hours
- Timeline: 25–33 days

2) Breakdown by workstream
… - Backend stabilization + DB hardening: 23–24h
- WebSockets + Redis + async tests: 28–32h
- Deterministic pricing engine + backfill: 22–24h
- Similar listings engine: 20–22h
- AI smart review + cache/versioning: 25–28h
- Notifications + catalog search + Facebook alignment: 18–20h
- Final QA, CI hardening, migration notes, docs: 14–15h

3) Risks / unknowns
- hidden placeholder logic may exist outside the obvious endpoints
- backfill may be needed once pricing/similarity become fully deterministic
- WebSocket behavior in multi-instance setup needs real integration validation
- AI evidence retrieval and cache invalidation rules need to be defined early
- schema drift between the core system and the Facebook module may require cleanup

4) Proposed milestone plan
- M1: audit + backend / DB stabilization
- M2: pricing + similarity engines
- M3: WebSockets + Redis hardening
- M4: AI review + cache + Facebook integration
- M5: tests / CI + final stabilization

5) Testing strategy
- deterministic unit tests for pricing and similarity rules
- async integration tests for WebSockets / Redis and the AI cache path
- DB integrity and backfill checks
- API schema contract tests
- regression pass on real sample data before handoff

6) Required access
- repo access
- full PDF brief
- PostgreSQL and Redis access or local docker-compose
- LLM / API keys
- Facebook test account or sample data
- staging environment if available

I noticed the previous project closed without being completed, so I recalculated the delivery time and adjusted the price accordingly.
If you think any additional information is needed from my side before we move forward, please let me know — I’ll be glad to help and answer any questions in private messages.

Hello.I have been working with Python/JavaScript for more than 8+ years.I'm ready to cooperate.

Hello, Max! I am Nina, the manager of developer Valentin. We have thoroughly reviewed your brief. The situation is clear: you have a "framework" that needs to be turned into a bulletproof system by removing randomness from pricing and ensuring the reliability of WebSockets.

Valentin specializes in AI-augmented development, which allows him to conduct deep refactoring and hardening of the architecture much faster than traditional teams, while maintaining a focus on determinism.

1. Preliminary estimate: ~95–110 hours

We aim for "no placeholders" quality, so we allocate time for complete test coverage and migrations.

2. Breakdown by workflows:
…
Pricing & Similarity Engines (25–30 hours): Implementation of tier logic, outlier filtering, weighted scoring, and data backfill scripts.

WebSocket & Redis (20–22 hours): Transition to asynchronous Redis Pub/Sub, socket mapping, support for multi-instance, and preference filtering.

AI Smart Review & Cache (18–20 hours): Generation of fingerprints, integration of evidence search, LLM synthesis, and caching with TTL.

DB Hardening & Search (15–18 hours): SQL fixes, indexes, integrity constraints, real filter search, and pagination.

Tests & CI (12–15 hours): Complete coverage through pytest-asyncio, stabilization of the CI pipeline.

FB Integration (5–8 hours): Connecting the existing module and resolving proxy issues.

3. Risks and uncertainties:

Data quality: Non-determinism in the past may have created "dirty" data that will require complex cleaning during backfill.

LLM Hallucinations: For Smart Review, strict prompt engineering will be required to ensure the AI does not go beyond web evidence.

WebSocket Scaling: With a sharp increase in the number of instances, atomicity of mutations must be ensured when iterating over sockets.

4. Milestone plan:

M1: Foundation: Stabilization of the database, indexes, removal of debug prints, implementation of real search.

M2: Determinism: Launch of new Pricing and Similarity engines + backfill.

M3: Communication: WebSocket system and integration with Facebook.

M4: Intelligence: Smart Review (AI) layer, caching, and final test coverage.

5. Testing approach:

We use TDD for calculation modules. First, we write tests for expected deterministic results of formulas, then we implement the logic. Integration asynchronous tests for WebSocket through mocks of Redis streams.

6. Required access:

Access to the repository (GitHub/GitLab).

Access to the staging environment (or Docker-compose for local launch).

API keys for LLM providers and test accounts for FB.

Show original

Translate

Hello! I have reviewed the project and am ready to start working. I am sure you will be satisfied with the result.

Show original

Translate

Your platform is already architecturally ready, but placeholders and non-deterministic logic in production are not just technical debt; they lead to incorrect pricing and inaccurate recommendations for real users. I specialize in such tasks: replacing placeholder logic with a full-fledged pricing engine with tiered selection of comparables, filtering outliers, and weighted scoring; implementing a WebSocket layer for real-time updates; connecting an AI review layer with deterministic validation rules. Approach: first, audit the current code and document contracts between modules, then iteratively replace stubs with test coverage to ensure nothing breaks. Estimated scope — 14 days, 950 USD. I am ready to discuss details after reviewing the full technical specification and repository.

Show original

Translate

We’re a team of system engineers and developers at SDEV, specializing in robust backend systems, production-grade APIs, and complex data workflows. We’ve reviewed your project in detail and are confident we can deliver a fully deterministic, reliable, and scalable version of your vehicle listings analytics platform.

Our approach will focus on:

- Replacing all placeholder logic with fully deterministic pricing benchmarks and similarity ranking, using tiered comparables, outlier detection, weighted scoring, and clear tie-breaking rules.
- Implementing a secure, per-user WebSocket notification system with Redis async pub/sub or streams, ensuring cross-instance reliability and proper authentication.
- Building the AI review layer end-to-end: real web evidence retrieval, LLM-based synthesis, and a fingerprint-keyed cache with TTL and version-based invalidation.
- Auditing and fixing SQL issues — missing constraints, indexes, and schema inconsistencies — with clean, versioned migrations.
- Delivering comprehensive async test coverage using pytest and pytest-asyncio, fully integrated into CI.
… - Connecting the Facebook integration module into the core system with proper error handling and monitoring.

We’ve handled similar systems involving real-time data, AI pipelines, and high-integrity analytics. A relevant case from our portfolio: Development of high-load analytics backend with real-time WebSocket updates, AI-driven insights, and PostgreSQL optimization — built on FastAPI, Redis, and async Python stack.

We propose a milestone-based delivery model aligned with your workstreams. Each milestone includes implementation, testing, documentation, and handover. We’ll provide a detailed technical breakdown, risk assessment, and access requirements upon confirmation.

Looking forward to collaborating.