Afinit

A Series E fintech company behind True Balance, a financial platform in India with over 100 million cumulative downloads.

Based on public company materials · Platform scale · Series E funding

Why I joined and what I worked on

Afinit’s work on financial access in India drew me to the company. I applied because I wanted to contribute to solving a problem that mattered beyond the software itself.

The AI transformation team’s goal was an environment where each employee could build and operate a service with AI—not just automate personal tasks, but create customer-facing services. I worked on the platform engineering behind that goal.

Shared development environment

I improved the shared harness in the company monorepo, including system prompts, skills, and hooks. Rather than waiting for an assigned task, I identified problems during use, submitted fixes, and incorporated them through PRs reviewed by our team lead.

For example, a skill was meant to enforce the PR template, but its instructions could be lost in a large context. I added a hook to reinforce compliance in the execution flow instead of relying solely on automatic skill selection.

Building and dogfooding the QA tool

I built a QA agent to validate employee-created services before deployment. Initially I focused on implementation, including a general-purpose VM-based execution environment, without narrowing down exactly what needed validation.

Using it at an internal hackathon with roughly 50 services showed that a polished implementation could still fit the problem poorly. Had web regression testing been the defined goal, I could have started with something smaller, such as Playwright and help authoring tests.

This experience changed my view of engineering: understanding the product problem and its usage context is part of the job, not something to leave outside implementation.

Proposing a direction for agent collaboration

While using a Slack-based agent collaboration environment, I proposed moving from human intervention at each step toward agents collaborating under human oversight. Human-on-the-loop became one of the team’s long-term directions.

I did not implement the agent collaboration system. My contribution was the direction proposed from hands-on use; evaluating its effectiveness and cost remained separate work.