Berapa peratus ticket deflection yang dicapai oleh AI customer support agent TTD?

73% ticket deflection dalam tempoh 6 minggu untuk deployment e-commerce — disertakan dengan architecture decision dan apa yang broke dalam production.

Berapa accuracy entity extraction untuk tender document processing ePerolehan?

91% entity extraction accuracy ke atas 977 government PDF tender documents — termasuk pengajaran mengenai kesukaran structured output dari dokumen kerajaan sebenar.

Apa pelajaran utama dari membina multi-agent system IRIS dari awal?

Enam bulan iterasi mendedahkan tiga kesilapan architecture yang costly — terutamanya berkaitan supervisor pattern dan state management yang betul untuk sistem multi-agent yang reliable.

Pillar 05 — Project Logs · Real builds · Real stacks · Real outcomes

← The Lab

Pillar 05 — Project Logs & Case Studies

Log Dari Lab: Project Builds & Case Studies

Ini bukan portfolio yang polish untuk impress. Ini adalah documentation sebenar — apa yang kami bina, decision yang kami buat dan kenapa, apa yang tak berkesan, dan nombor sebenar bila kami boleh share.

3 logs published

BOFU — Decision stage

All technical levels

Production data active

◈

Publishing policy untuk Pillar 5: Kami tak publish log sampai ada data sebenar untuk share — timeline, stack, dan outcome metric yang honest. Kalau ada NDA atau client sensitivity, kami anonymize industry tapi kekalkan technical detail. Logs akan keluar bila IRIS v2.0 dah cukup production time, atau bila client projects dah boleh di-disclosed.

Log #001

Customer Support Agent · Stack: Claude + LangGraph + FastAPI

Building a Customer Support Agent from Scratch — Architecture Decisions & What We'd Do Differently

Full walkthrough dari requirements gathering sampai production deployment. RAG setup, escalation logic, HITL approval flow, dan real performance numbers.

Case Study Project Log Technical

New Log #002

Tender Document Processing · Stack: Claude + pgvector + FastAPI

Tender Document Processing Pipeline — Extracting Structured Data from Unstructured Government PDFs

ePerolehan document parsing, entity extraction, compliance flag detection. Masa processing sebelum vs selepas. Accuracy benchmarks dan failure modes kami encounter.

Case Study Project Log Procurement

New Log #003

Multi-Agent System · Stack: LangGraph + IRIS v2.0

Building a Multi-Agent System from Scratch — 6 Months of Iteration on IRIS

The honest story behind IRIS v2.0 — what the v1 architecture got wrong, the supervisor-specialist pivot, state management decisions, and what production taught us that design reviews didn't.

Case Study Project Log Technical

New

Want the specifics of these builds?

Kalau korang evaluating something similar dan nak discuss architecture atau lessons learned secara direct — book a session. Kami boleh share more in context of your specific use case.

Book Teh Tarik Session ❯❯

LOGS IN THIS PILLAR

WRITTEN FOR

Engineers evaluating similar builds
Founders making AI investment decisions
CTOs assessing build vs buy vs partner
Anyone who's tired of sanitised case studies

NOTIFY ME

Get notified when new insights are published.

Log Dari Lab: Project Builds & Case Studies

Want the specifics of these builds?

LOGS IN THIS PILLAR

RELATED PILLARS

WRITTEN FOR

NOTIFY ME