Roast My TechStack #5 Recap

Speaker 01

Agents To Find Business Opportunities

Richard Lee

How he built a multiple-agent system to hunt for business opportunities — an executive agent controlling many siloed, purpose-built child agents. The deck is available on GitHub along with the painful lessons about grounding and over-engineering.

Multi-agent · Orchestration · Grounding · Containerised auth

LinkedIn GitHub

One Executive, Many Child Agents

Multiple agents work together: an executive agent controls many child agents. The critical design rule — a child agent must be very siloed and purpose-built. Each one does one job well and stays out of the others' business.

Hallucination Requires Layered Grounding

The key challenge was hallucination, and the answer was multiple layers of grounding — no single grounding source is enough when agents are compounding their own context.

Don't Over-Engineer

Don't jump straight into micro-services — a monolith might be better to start. You can always transition to scale later, and speed is still important in the early days of an agent system.

Deterministic Where It's Destructive

If an operation is high-risk or destructive, always go for the deterministic approach (code) rather than an agent. Never roll your own auth — airlock your auth by containerising it.

For observability, consider Grafana.

Speaker 02

PixelPuggles

Alec Teo

An image app that pixelates and animates photos — roasted on its auth choice, its data schema, and how it can be made agent-friendly. The room also weighed image-generation cost and automating image evaluation.

Supabase auth · Clerk · Data schema · Agents-ready · Imagen

Supabase Auth vs. Clerk

The choice: create an auth after 3 plays for lower barrier of entry, natively integrated with supabase. The trade-off — Supabase auth's messaging isn't as great, but Clerk's uptime isn't as good as Supabase.

Review The Schema Yourself

Data schema is still very important in system design, and you need to review it yourself — don't let the framework decide it for you.

Get Agent-Ready

Allow agent interaction with your app to prep for an agentic future. Check your surfaces against isitagentready.com to make your app more agent friendly.

Image Processing & Cost

Alec manually pixelated the photos and batch-uploaded them. Consider offloading pixelation to the user's device to better optimise. For image generation costing, consider Imagen 4 (6× cheaper) over competing generation models.

Automate Image Eval

Consider automating part of the image generation evaluation: use an LLM as a judge to check output. Two rules from the room: it must be a different model from the image gen model, and it should have access to a scientific database for best accuracy.

Speaker 03

Evokoa · pgGraph

Dalton Prescott Ng

A virtual graph layer for AI agents built on Postgres — pgGraph. The session covered why graph databases are expensive, when graph beats RAG, and how database hydration unlocks use cases like fraud detection.

Graph DB · PostgreSQL · RAG vs Graph · Hydration

LinkedIn GitHub

Graph Databases Are Expensive

Graph databases carry a real cost. pgGraph's approach on Postgres is roughly 34× cheaper than Neo4j, bringing graph capabilities into a database teams already run.

Database Hydration

Database hydration (also known as object hydration) converts raw database query results — a row or an array — into fully functioning, in-memory programming objects. A big use case is fraud detection.

RAG vs. Graph

The room's clean takeaway: RAG is good for search; Graph finds the relations between entities. Use the right question per tool instead of forcing one to do both.

Venue sponsor

SGInnovate

PIC: Arul · Co-hosted with GrowthLab

Event ops volunteers

⏳ Time & Mic Keeper

Ying Zhao

📸 Memories Keeper

Esther

🎙 Host

Jesse

🧠 Knowledge Keeper

Han

Roast My TechStack

Original published recap ↗