Case studies

Five production agents
we've shipped.

Real deployments across requirement engineering, CI/CD, test orchestration, B2B trade, and career pipelines. Three live in production at a leading Bangladeshi enterprise software company; two we built for our own operations and continue to maintain.

PRODUCTION · SINCE MAY 2026

Echo — agentic requirement engineering

Cross-channel requirement drift was costing deliverables. Clients raised needs across calls, chats, emails, and shared files; each developer heard only their slice. Nothing was reconciled.

What we built:

  • Auto-captures from 4 channels: calls, chats, emails, files. Bangla speech → English in one transcription pass.
  • One nightly AI pass identifies real client requirements, drafts one verification email per client.
  • Human reviews + sends; approved items auto-populate the backlog tracker.

Status

Live in production

Net new cost

$0

PRODUCTION · RUNNING NIGHTLY

Nightwatch — multi-layer test orchestration with AI triage

Enterprise products need load, API, integration, and E2E coverage — but each layer lives in its own dashboard. Failures get lost in the noise.

What we built:

  • One agent orchestrates four test layers nightly at 23:59 BDT: Locust load, Newman API, pytest integration, Playwright E2E.
  • Claude analyzes failures — real defect vs flake vs environment vs known-issue; identifies trends across nights.
  • One consolidated PDF report emailed to leadership by 02:00 BDT every morning. No dashboards required.

Test layers

Load · API · Int · E2E

Daily delivery

02:00 BDT

PRODUCTION · SINCE APRIL 2026

Forge — agentic CI/CD with auto-rollback and bug routing

Most teams treat CI/CD as a one-way street: build, deploy, hope. When things break post-deploy, rollback needs human coordination, bugs get filed late, and the team learns about failures from the support inbox — not from the pipeline.

What we built:

  • On every push: pulls latest code, snapshots the current deploy, builds, deploys, runs the full E2E suite against the new build.
  • Green path → keeps it. Red path → auto-rollback to the snapshot within seconds; production is safe before anyone notices.
  • Files a structured bug ticket (failing test, commit hash, artifacts) and emails engineering leadership + the team with the outcome — every time.

Rollback time

Seconds

Stack

Claude SDK + Playwright

PRODUCTION · LIVE AT SATUMM

SATUMM-Agent — cross-border B2B trade discovery & outreach

Bangladesh has world-class production capacity, but cross-border buyer discovery is manual and slow. International importers in MY/US/AU lack vetted Bangladeshi supplier networks.

What we built:

  • Discovers buyers / importers / exporters across Malaysia, USA, Australia using public business registries.
  • Drafts personalized cooperation proposals per target — not template spam. CAN-SPAM / Spam Act / PDPA compliant.
  • Tracks every interaction in Google Sheets funnel. Nightly summary email of what was done.

Geographies

MY · US · AU

Stack

Claude SDK + Sheets

ACTIVE · IN-HOUSE TOOL

Digital-Deputy — AI-augmented career pipeline

Career pursuit is a multi-channel pipeline: research → tailor → outreach → follow-up → track. Done well it's a sales process; done poorly opportunities slip. Most do it poorly because overhead is enormous.

What we built:

  • Sources targeted opportunities. Generates tailored resumes per opportunity (not spray-and-pray) via a programmatic resume generator.
  • Drafts personalized outreach from two SMTP identities (personal + business); polls IMAP for replies.
  • Sheets funnel + durable agent memory across sessions; indexes past Claude Code sessions for context recall.

Identities

2 SMTP + IMAP

Stack

Claude CLI + Sheets

Want one of these for your team?

See services →