Agentic AI Workflow Design
Multi-step orchestration, tool calling, fallback logic, and human-in-the-loop design.
AI Solutions → Product Management
Merlin Treesa Benny.
Moving into product management, one shipped AI feature at a time.
At Phronetic AI, I work with enterprise clients to understand their problems, write requirements, and improve AI agents. As an intern, I proposed a platform feature that shipped, improving accuracy by 20% and cutting annotation time from 1 hour to 40 minutes.
The work
What I do
Requirements → Shipped Product
Specs, user flows, acceptance criteria, and end-to-end delivery ownership.
Client Embedding
Working inside enterprise operations to find where automation creates real value.
Evaluation & Iteration
Systematic prompt evaluation, failure-pattern analysis, and accuracy improvement cycles.
Proof, not promises
Highlights
20%
Accuracy improvement from shipped feature
33%
Faster annotation (1 hr → 40 min)
35%
Agent accuracy improvement
3+
Enterprise clients (retail, fintech, healthcare)
Selected work
Featured Projects
Retail KPI Monitoring Agent
Problem
Teams manually reviewed dashboards every day to catch KPI misses, so risks were often spotted too late.
User
Retail operations teams and store managers.
Key decision
Instead of another dashboard, built an agent that compares daily actuals against targets, classifies risk, and pushes Slack alerts, with email reserved for high-risk cases.
Result
Risks are surfaced through daily checks and routed by severity, reducing reliance on manual dashboard review.
GPT-4.1Zoho AnalyticsSlackEmailTool Calling
View project →Customer Support Agent
Problem
Support queries needed more consistent routing and resolution across different customer intents.
User
End customers and the support team.
Key decision
Designed intent detection, entity extraction, tool calling, and fallback logic, with clear rules for when to hand off instead of guessing.
Result
Improved agent accuracy by 35% through prompt evaluation, test-case creation, failure-pattern analysis, and root-cause analysis.
PythonPhronetic AI PlatformPostman (API Testing)
View project →Where Is My Train — Product Case Study
Problem
How did Sigmoid Labs build and scale a consumer app, and what should it do after acquisition?
User
Product teams evaluating the app’s next phase of growth.
Key decision
Applied Atlassian’s Product Lifecycle framework to user segmentation, competitive positioning, business model, and key trade-offs.
Result
Produced product recommendations for the post-acquisition phase.
Notion
View project →