The network for creativity
Join 1.25M professional creatives like you
Connect with clients, get discovered, and run your business 100% commission-free
Creatives on Contra have earned over $150M and we are just getting started
Back to feedPost
If you cannot audit an agent, you cannot trust it.
Production agents need a trail. What input arrived. What tool ran. What got written. Who approved the risky step.
Without that, debugging turns into guesswork and clients lose trust fast.
I design AgentOps the same way I design CRM sync. Explicit steps. Retries. Logs. Human review where it matters.
Message me if you want an agent that is usable in production, not only in a demo.
The network for creativity
Join 1.25M professional creatives like you
Connect with clients, get discovered, and run your business 100% commission-free
Creatives on Contra have earned over $150M and we are just getting started
Related posts
Building AI agents is easy. Building one you can trust with a real business workflow is harder.
I built a Freelance Admin Agent to explore that problem using Strands Agents SDK and Claude Sonnet 4.5 on Amazon Bedrock.
The working MVP takes manually provided client emails and turns them into a structured payment workflow:
→ Classifies invoice and payment emails
→ Extracts client, invoice, amount, currency, dates and contextual details using structured output
→ Stores and matches records in SQLite
→ Uses deterministic logic to identify full/short and on-time/late payments
→ Uses a second agent to draft a contextual follow-up only when needed
→ Keeps the final decision with the human through Approve / Edit / Reject
One of the most valuable parts wasn't adding another AI feature. It was finding a real reliability issue during testing: reprocessing the same email could duplicate records and inflate payment totals.
I reproduced it through repeat testing, traced the database behavior, and fixed it.
That reinforced something I'm increasingly applying to my AI engineering work:
A useful AI product isn't just an LLM call. It needs clear boundaries between AI reasoning, structured data, deterministic business logic, persistence, testing, and human control.
The next iteration will focus on stronger production safeguards, including source-grounded financial extraction, precise currency handling, stricter trust boundaries for untrusted input, and eventually live inbox integrations.
I'm interested in building more systems like this for startups and teams with repetitive workflows that can genuinely benefit from AI agents and automation.
Stack: Python · Strands Agents SDK · Amazon Bedrock · Claude Sonnet 4.5 · Pydantic · SQLite · Streamlit
turning client emails into admin tasks is such a real freelancer pain, nice build 🔥
Designed and built an AI operations dashboard that brings projects, decisions, tasks, research, connected tools, and voice interaction into one clear workspace. The goal is to turn complex workflows into a simple, usable system that helps people see what matters and act faster.
Seeing the voice interaction layer pull data from Supabase into the dashboard cuts down on manual context switching.
I built PREBUYR in Lovable for the @Lovable challenge — a route-aware booking system for mobile pre-purchase vehicle inspections.
The problem: a buyer wants the car inspected, the seller controls when the vehicle is available, and the inspector is already moving between appointments. A normal booking calendar can show a “free” slot that the inspector physically can’t reach.
PREBUYR turns that three-way coordination into one flow:
→ The buyer requests an inspection
→ The seller shares when the car is available
→ PREBUYR checks the inspector’s existing route, travel time and operating buffer
→ The seller sees only reachable appointment times
→ Once confirmed, the buyer and inspector update automatically
→ Rescheduling keeps the same appointment in sync instead of creating a duplicate
The key idea is simple:
Free doesn’t always mean reachable.
The current demo is built around a solo inspector serving the Denver metro area, but the same coordination layer can scale much further — from one inspector, to a multi-inspector company with multiple service areas, and eventually to a marketplace where customers can choose between specialists while the system coordinates availability and routing behind the scenes.
The goal is to remove as much manual scheduling and back-and-forth from the business owner as possible:
From “Can I book?” to “You’re booked.”
Partner certification note: in accordance with the challenge requirements, I submitted an application for the Lovable Partner certification. However, I’m not currently eligible to receive the certificate because I only started using Lovable at the beginning of this challenge, while the certification requirements include more than 3 months of subscription history and more than 3 published applications.
