IMDA OIP Call 28 — NTUC LHub

The system
carries
the load.

COMPASS is an intelligent, memory-driven platform that automates career placement programmes end-to-end — so every human involved can focus entirely on work only humans can do.

Explore the product

WATCH: COMPASS IN 3 MINUTES

10–30% Improved placement rate target
50% Reduction in admin time target
5 Personas served end-to-end
93% Memory retrieval accuracy (LoCoMo benchmark)

Experience without memory. Process without intelligence.

Every year NTUC LHub runs programmes that change careers. The people running them are committed. The problem is the system they work within makes every part of their job harder than it needs to be.

These are not four separate problems. They are one problem expressed four ways: a system that generates experience without capturing it, accumulates data without learning from it, and places humans in the role of connective tissue that the software should be providing.

The failure cascade
01 Broken enrolment — subjective assessment, inconsistent decisions, wrong trainees in wrong courses
02 Unplaceable graduates — training completed, skills mismatched, employers frustrated
03 Employer distrust — bad matches erode confidence, relationships deteriorate
04 Staff firefighting — coordinators patch holes manually, capacity consumed
05 No early intervention — by the time the signal is visible, the opportunity to act has passed
📋

Enrolment by intuition

Salespersons manually assess suitability with no outcome data. Inconsistency baked in from day one.

👻

Trainees go silent

Post-training follow-up by email. No engagement, no response. Data collection fails at the moment it matters most.

🔀

Matching done by hand

CVs collected manually. Employer matching done by humans with no systematic memory of what worked before.

📊

Reports under pressure

Government reports compiled manually from Excel. Errors, delays, and zero time to intervene before submission.

Five people. One broken system.
One solution that serves them all.

👩‍💼

Sarah

Programme Coordinator

Manages 200+ active trainees daily across multiple programmes. Spends hours switching between CRM, SharePoint and email.

What COMPASS gives her

A daily brief of decisions that need a human. Everything else — follow-ups, reminders, status updates — runs autonomously. Her cognitive load drops by 80%.

🧑‍💻

James

Enrolment Salesperson

First touchpoint for every trainee. Currently the inconsistency bottleneck — assessing suitability manually with no system support.

What COMPASS gives him

A decision brief for every application: outcome-weighted recommendations with full reasoning drawn from thousands of prior trajectories. His job becomes approval and relationship.

🙋

Marcus

Trainee / Career Switcher

35-year-old retail manager transitioning to ICT. Completed training. Currently receiving generic email reminders he ignores.

What COMPASS gives him

A live placement feed showing employer views, match scores, and specific skill gaps. The system delivers value before asking for anything. Engagement becomes rational.

🏢

Alex

Employer / Hiring Manager

Receives candidate referrals from NTUC LHub. Currently frustrated by poor match quality and reactive placement process.

What COMPASS gives him

A live pipeline of candidates currently in training who match his profile — before they're available. The system learns his revealed preferences over time. Bad matches become rare.

📑

Lin

Compliance Officer

Prepares government reports for SSG monthly. Currently compiling manually from Excel under deadline pressure.

What COMPASS gives her

Reports that write themselves. A continuous compliance state updated in real time. One-action submission. Her job becomes certification, not compilation.

Marcus's journey — from application to placement

The most critical and most neglected journey in the system

📝
Applies
Submits application. AI enrolment agent analyses background and career goals
🎯
Matched
Outcome-weighted course recommendation. Salesperson reviews and confirms
📚
Trains
Profile updated continuously. Employer pipeline starts forming before graduation
👁️
Visible
Live feed: employer views, match scores, skill gaps. Not a status page — a career companion
Placed
Employment verified via MyInfo. Outcome feeds back into enrolment intelligence
📈
Tracked
90-day, 180-day retention check-ins. System learns what produces durable placements

Three interfaces. One intelligence layer.

COMPASS — Interactive Prototype
Coordinator Brief
Placement Matching
Trainee Feed
Good morning, Sarah
Friday, 17 April 2026 — 3 actions need you today
3 actions
⚠️
Marcus Tan — Placement at risk
No employer response in 14 days · Match score dropped to 61%
🎯
Deloitte — Pipeline gap detected
3 trainees within 2 skills of qualifying · Suggested: fast-track module
SSG Monthly Report — Ready for review
Auto-compiled · 94 placements · Submission deadline in 2 days
NATURAL LANGUAGE QUERY
Show me all trainees within 2 skills of the Deloitte BA opening
↳ Found 3 trainees · Avg gap: 1.3 skills · Est. ready: 3 weeks
SHOWING 4 CANDIDATES — RANKED BY PLACEMENT PROBABILITY
91%
Priya K. — Retail Operations Manager, 8yr
BA Programme · Available Jun 2026 · 0 skill gaps
84%
Marcus T. — Sales Manager, 6yr
BA Programme · Available May 2026 · 1 skill gap: SQL
71%
Wei L. — Project Coordinator, 4yr
BA Programme · Available Jul 2026 · 2 skill gaps
COMPASS INSIGHT — Candidates with retail ops background placed at 2.3× rate for BA roles vs technical backgrounds at this employer type
M
Marcus Tan
BA Programme · Week 8 of 12
ON TRACK
👁️
Accenture viewed your profile
2 hours ago · Business Analyst role · Match: 78%
🎯
New match: Deloitte Digital — BA Consultant
Match score: 84% · 1 skill gap remaining
📈
3 employers have your profile in active consideration
Up from 1 last week
Your skill readiness for Deloitte BA role
Business Analysis
92%
Stakeholder Mgmt
88%
SQL / Data
45%
Agile Methods
79%

Built to learn. Designed to scale.

Hover over any component to understand its role, its technology, and why it was designed that way.

Frontend — Three purpose-built interfaces
🖥️
Coordinator Portal
Next.js 14
📱
Trainee Portal
Next.js PWA
🏢
Employer Portal
Next.js
API & Event Layer
FastAPI Backend
Python · Async
🔀
Event Bus
Redpanda / Kafka
🔐
Auth / Authz
Keycloak · OPA
Intelligence Layer — The architectural heart
🧠
Agent Orchestrator
LangChain
📋
Enrolment Agent
Claude API
💬
Engagement Agent
Multi-channel
🎯
Matching Agent
Weaviate · Semantic
📊
Compliance Agent
Rule-based + LLM
Memory Layer — Three complementary stores
🔮
Vector Store
Weaviate
🕸️
Graph Store
Neo4j
📁
Document Store
MongoDB
🗄️
Operational DB
PostgreSQL
Cache / Queue
Redis

Every choice argued.
No fashionable defaults.

Backend Runtime

Python + FastAPI

vs Node.js, Go

The AI ecosystem lives in Python. Keeping the backend in the same language as the intelligence layer eliminates serialisation overhead for every agent call. FastAPI gives async performance comparable to Node.js for I/O-bound workloads.

Python / FastAPI Node.js Go
Vector Store

Weaviate

vs Pinecone, pgvector

Open source, self-hostable, with hybrid search combining semantic vectors and keyword matching in one query. Pure semantic search misses exact credential matches. Pure keyword misses semantic similarity. Weaviate handles both.

Weaviate Pinecone pgvector
Graph Store

Neo4j Community

vs SQL joins, Apache AGE

Outcome pathway intelligence is a graph problem. Which course feeds which employer type, which profile trajectories produce durable placements — Cypher traversals answer these elegantly where SQL joins across five tables fail.

Neo4j Apache AGE SQL
LLM Layer

Claude API + Ollama

vs OpenAI-only, proprietary platforms

Frontier reasoning for complex assessment. Local Ollama models (Llama 3 / Mistral) for high-frequency routine tasks at zero marginal cost. LLM layer is abstracted — switching providers is a config change, not a rebuild.

Claude + Ollama OpenAI only Azure OAI
Event Streaming

Redpanda

vs Kafka, RabbitMQ, Redis Pub/Sub

Kafka-compatible API, single binary deployment, no ZooKeeper. All the architectural benefits of Kafka without a dedicated ops specialist. RabbitMQ loses event replay. Redis Pub/Sub loses messages on consumer downtime.

Redpanda Kafka RabbitMQ
Agent Orchestration

LangChain

vs custom, LlamaIndex

Tool calling, memory injection, chain composition — solved problems. Building this from scratch has no strategic value. LlamaIndex is better for document RAG specifically; LangChain wins on general multi-agent orchestration.

LangChain LlamaIndex Custom

Not better features.
Better thinking.

01

Memory that compounds

Competing solutions store data. COMPASS builds knowledge. Every outcome feeds back into the next decision. The system widens its advantage over time — not just at launch. No competitor has closed the feedback loop between placement outcomes and enrolment decisions.

02

Agents that act, not alerts that notify

Most platforms surface information and wait for humans to respond. COMPASS deploys agents that monitor, act autonomously on routine tasks, and surface only decisions that require judgment. The cognitive load reduction is structural, not incremental.

03

Enrolment intelligence closes the root cause

Competitors address matching at placement stage — after the damage is done. COMPASS addresses it at enrolment, where poor decisions cascade into every downstream failure. This is the highest-leverage intervention in the entire system.

04

Trainees as stakeholders, not data subjects

The ghosting problem is a design failure, not a communication failure. COMPASS delivers visible value before asking for anything. Engagement becomes rational. The system earns participation rather than extracting it through reminder volume.

05

Open architecture. No lock-in.

Entire stack is open source. No per-seat licences. No proprietary platform fees. LLM layer is abstracted — switching providers is a config change. The cost of scaling is compute and data volume, not vendor tiers.

06

Demonstrated in production-adjacent systems

GridTrace and Cassandra.ai prove the architectural patterns work. Institutional memory agents that learn from outcomes and improve with every session. This is not theoretical — it has been built, demonstrated, and validated.

Stress-tested.
Not just designed.

LLM non-determinism in regulated decisions

HIGH IMPACT

Two identical applications submitted an hour apart should not receive different course recommendations. In a regulated programme, non-determinism is a compliance liability.

Mitigation

Temperature set to zero for assessment tasks. Every decision logged with full prompt, context, model version, and output. Recommendations frozen at decision time. Human sign-off required before any recommendation becomes an action.

Trainee adoption failure

HIGH IMPACT

If trainees don't engage, the memory layer stays thin, matching quality stays low, and placement outcomes don't improve. The entire system depends on data only trainees can provide.

Mitigation

The portal delivers visible value before asking for anything. First session shows live placement pipeline activity. WhatsApp-first reduces portal dependency. The system gives before it takes.

Internal adoption — perceived surveillance

HIGH IMPACT

COMPASS tracks coordinator interventions and salesperson override rates. If staff perceive this as performance monitoring, adoption fails regardless of technical quality.

Mitigation

Staff behaviour data used exclusively to improve system recommendations, never for performance management. This commitment is contractual and visible in the product itself.

Multi-store data consistency

MEDIUM IMPACT

Four stores with no native transaction boundary. A partial write — profile updated in PostgreSQL but embedding stale in Weaviate — means the matching agent reasons from outdated data.

Mitigation

Saga pattern via the event bus. Each store update is an idempotent event consumer. Nightly reconciliation job compares checksums across stores and flags drift.

Full architecture is the vision. Prototype proves the thesis.

The three-store memory architecture, the full agent suite, and the MyInfo integration represent the production system. For the prototype, a single-store implementation (PostgreSQL + pgvector) demonstrates the intelligence layer end-to-end. This is honest and defensible — evaluators are assessing the thinking. The prototype shows the thinking works. The architecture shows what it becomes.

The system gets smarter every day it operates.

COMPASS is ready to build. The architecture is defined. The technology is chosen. The risks are mapped. The prototype scope is clear. What's needed is selection — and the opportunity to build the intelligence infrastructure Singapore's workforce development ecosystem has needed for years.

"Not selecting this is not a conservative choice — it is choosing a system that will automate today's processes and stagnate. COMPASS is the only submission that will still be getting better in year three."