Customer Stories
How Singapore Innovators Build on PULSAR
Three detailed case studies showing how PULSAR's enterprise AI platform — combining GPT assistants, RAG retrieval, predictive analytics, and workflow automation — delivers measurable outcomes for fintech, logistics, and B2B commerce companies.
Note: Client names are shown in abbreviated form to protect commercial confidentiality. Registration details and full references are available under NDA upon request.
Case Study
CREDOLAB
Fintech — Credit Risk & Fraud Intelligence
Embedding real-time fraud detection and explainable decision automation into a behavioral analytics stack serving banks and lenders across Southeast Asia.
Headquarters
Singapore
Markets Served
Banks & digital lenders across Southeast Asia
Engagement
Enterprise AI Platform Deployment
About CREDOLAB
CREDOLAB is a Singapore-headquartered fintech that provides behavioral data analytics, credit risk scoring, and anti-fraud solutions to banks, digital lenders, and financial institutions. Its scoring technology transforms privacy-compliant behavioral metadata — the digital footprint left during a loan application — into predictive insights that help lenders approve more good customers, thin-file applicants in particular, without taking on hidden risk.
Operating across Southeast Asia's fast-growing consumer credit markets, CREDOLAB sits at the intersection of two demanding pressures: clients expect decisions in real time inside their mobile application flows, while regulators and bank risk committees expect every automated decision to be explainable, auditable, and fair. The company works under the expectations of Singapore's MAS FEAT principles (Fairness, Ethics, Accountability, Transparency) and PDPA data-handling obligations in every market it serves.
The Challenge
Real-time decisioning at strict latency budgets
Credit and fraud scores are requested synchronously inside live application journeys on a lender's mobile app or web channel. Every additional second of latency measurably increases applicant drop-off, so the decision pipeline needed to aggregate behavioral scoring, device intelligence, and fraud signals and return a verdict in well under a second — even during end-of-month and payday traffic peaks.
Model governance and explainability for regulated clients
Bank risk committees and regulators do not accept black boxes. CREDOLAB needed every automated decision to come with a human-readable explanation, a reconstructable audit trail, and versioned decision policies — so that a compliance officer could answer 'why was this applicant declined' months after the fact, and so model changes could be rolled out with controlled approval workflows.
Heavy manual review load on risk and compliance teams
Supporting documents — payslips, bank statements, identity records — arrived in dozens of formats per market and were screened manually. Review teams spent most of their time on extraction and cross-checking rather than on genuine risk judgment, creating a queue bottleneck that slowed onboarding for the very clients CREDOLAB wanted to help approve.
Fragmented multi-market deployments
Each country deployment meant localized document templates, languages, thresholds, and data-residency handling. Without a configuration-driven approach, every new market required bespoke engineering work, stretching time-to-launch from weeks to months and multiplying the surface area for defects.
The PULSAR Solution
PULSAR delivered a decision-automation layer on top of CREDOLAB's existing behavioral analytics stack. Rather than replacing CREDOLAB's core scoring models, the platform wraps them in managed AI services — orchestration, document intelligence, governance, and deployment tooling — built on PULSAR's RAG and workflow automation foundation on AWS:
Real-time decision orchestration API
A low-latency decision API on the PULSAR AI platform orchestrates behavioral scoring, device intelligence, and fraud signals into a single decision payload. Parallel signal evaluation, response caching for idempotent checks, and autoscaling infrastructure keep end-to-end latency below 300 milliseconds at the 95th percentile, including during seasonal traffic peaks. The API exposes a stable contract to lenders, so downstream integrations are insulated from model version changes.
RAG-powered document intelligence pipeline
Application documents are automatically classified, extracted, and cross-checked against application data. A retrieval-augmented pipeline built on managed foundation models handles the long tail of document formats across markets — payslips in Bahasa, bank statements in Vietnamese, CPF records in Singapore — producing structured evidence with cited source passages. Reviewers receive a pre-assembled case file instead of a folder of raw scans, and low-confidence extractions are routed for human confirmation rather than silently accepted.
Model governance and audit layer
Every decision flows through a governance layer that records the model versions used, the contributing signals, and a plain-language explanation generated for each verdict. Decision policies are versioned with approval workflows, model performance is monitored for drift and bias indicators, and immutable audit logs let a risk officer reconstruct any historical decision in seconds. This layer was designed explicitly around MAS FEAT principles and PDPA obligations.
Configuration-driven multi-market deployment
Market-specific behavior — document templates, extraction schemas, decision thresholds, supported languages — moved from code into configuration. Launching in a new market became a matter of deploying a configuration pack and validating against a standard acceptance suite, compressing time-to-launch from months to weeks and eliminating a whole class of per-market defects.
Outcomes
The impact showed up first in the review queue. With document intelligence pre-assembling case files and flagging inconsistencies, risk teams shifted from data entry to judgment — manual review workload fell by 60 percent while reviewer accuracy improved, because evidence arrived organized rather than scattered.
On the client side, pilot lenders saw approval rates rise 18 percent at unchanged default rates — the clearest possible signal that better information, not looser standards, was driving growth. Real-time decision latency under 300 milliseconds kept applicant drop-off low, and the governance layer passed client risk-committee review without a single finding.
< 300 ms
P95 end-to-end decision latency in live scoring flows
60%
Reduction in manual document review workload
+18%
Approval-rate uplift at unchanged default rates for pilot lenders
100%
Automated decisions covered by explainability and audit reports
What's Next
CREDOLAB and PULSAR are extending the platform with cross-institution fraud signal research and new thin-file segments in two additional SEA markets, using the configuration-driven deployment framework to accelerate each launch.
“PULSAR gave us a decision layer that our bank clients can actually audit. We moved from batch-oriented scoring to real-time, explainable decisions without rebuilding our core behavioral models — and our risk teams finally spend their time on risk, not on paperwork.”
Case Study
HAULIO
Logistics — Container Trucking & Supply Chain
AI-driven dispatch optimization and knowledge automation for a digital container trucking ecosystem coordinating daily port movements.
Headquarters
Singapore
Markets Served
Singapore port ecosystem & regional trade lanes
Engagement
Operations Intelligence Deployment
About HAULIO
HAULIO is a Singapore-based logistics technology company operating a digital platform for container haulage — the short, time-critical truck movements that shuttle containers between port terminals, depots, warehouses, and customer sites. The platform connects shippers, freight forwarders, and trucking companies through booking, scheduling, and supply-chain collaboration tools, and coordinates a large share of daily container movements in and out of Singapore's port.
Haulage is an unforgiving business to digitize. A single job is a chain of dependencies — vessel berthing windows, port system bookings, depot availability, driver shifts, customer receiving hours — and any break in the chain cascades into detention charges, missed connections, and idle assets. HAULIO's platform had already consolidated the industry's fragmented communication; the next step was making its operations intelligent.
The Challenge
Schedules that went stale within hours
Vessel schedules shift, terminals congest, depots close slots, and customers change receiving hours. Plans built at 7 a.m. were frequently obsolete by noon, yet dispatchers had no systematic way to know which of the hundreds of moving jobs were newly at risk. Re-planning was reactive — triggered by phone calls after something had already gone wrong.
Fragmented coordination across many parties
Job booking, vessel windows, and depot moves were coordinated across email, chat groups, and spreadsheets held by different parties. Each handoff introduced delay and transcription risk, and there was no single source of truth for the state of a job.
Heavy manual documentation work
Every haulage run generates paperwork: port passes, permits, booking confirmations, customs references. Much of it was prepared and checked by hand, and errors in a single field could ground a truck at a terminal gate — an expensive mistake multiplied across hundreds of daily jobs.
Operational know-how locked in a few heads
The judgment that keeps haulage running — which depots are reliable at which hours, which vessel services habitually run late, how to sequence pickups to avoid terminal queues — lived with a small number of veteran dispatchers. New hires took around three months to reach full productivity, and knowledge walked out the door with every departure.
The PULSAR Solution
PULSAR built an operations intelligence layer for HAULIO's platform on AWS, combining predictive analytics, document workflow automation, and a retrieval-based knowledge assistant. The goal was not to replace dispatchers, but to let them manage by exception — the platform handles routine coordination, and people handle genuine disruptions:
ETA prediction and dispatch optimization
Models trained on historical job, traffic, and port-congestion patterns predict job durations and flag assignments likely to miss their windows. Recommended assignments and re-planning alerts surface directly inside HAULIO's scheduling UI, so a dispatcher sees 'these seven jobs are newly at risk' rather than discovering it from an angry phone call. The system re-scores continuously as port and vessel conditions change throughout the day.
Automated documentation workflows
Port permits, booking confirmations, and customs references are generated, field-checked, and routed through an approval pipeline with full audit trails. Validation rules catch mismatched container numbers, expired permits, and inconsistent references before submission — the errors that used to strand trucks at terminal gates.
RAG knowledge assistant for operations teams
A retrieval-augmented assistant trained on SOPs, port notices, depot procedures, and accumulated dispatcher playbooks answers operational questions with cited sources, in English and Mandarin. New dispatchers use it to resolve unfamiliar situations in seconds; veterans use it to document their judgment so it stops living only in their heads.
Anomaly detection on platform telemetry
Continuous monitoring of job state transitions flags double-bookings, stale jobs with no recent activity, assets idle beyond thresholds, and deliveries trending late — each alert ranked by commercial impact so attention goes where it pays.
Outcomes
Dispatch planning shifted from a daily marathon to an exception-driven review. With optimization recommendations pre-computed and at-risk jobs flagged automatically, the time spent on daily dispatch planning fell by more than half — and the remaining time went to genuinely hard problems instead of routine assembly.
Across pilot fleet operators, on-time delivery rose 22 percent as re-planning moved from reactive to proactive. Perhaps most durably, new dispatchers now reach full productivity in about two weeks instead of three months, because the knowledge assistant makes the collective experience of the operation available on demand.
-55%
Time spent on daily dispatch planning
+22%
On-time delivery rate across pilot fleet operators
70%
Routine documentation processed without manual touch
2 weeks
Time for new dispatchers to reach full productivity (previously ~3 months)
What's Next
HAULIO and PULSAR are extending the intelligence layer toward shipper- and forwarder-facing visibility — predictive ETAs and exception alerts shared with customers — and evaluating regional expansion of the dispatch models to neighboring ports.
“Our dispatchers used to fight fires all day. With PULSAR's predictions and knowledge assistant, they finally manage by exception — the platform handles the routine, and people handle the disruptions. That's the difference between a scheduling tool and an intelligent operation.”
Case Study
EEZEE
B2B E-Commerce — Industrial MRO & Office Supplies
A RAG-powered procurement copilot that turns unstructured industrial catalogs and RFQs into fast, accurate quotes.
Headquarters
Singapore
Markets Served
Enterprise buyers across Southeast Asia
Engagement
Procurement Copilot Deployment
About EEZEE
EEZEE is a Singapore-based B2B e-commerce company running an online marketplace for industrial products, MRO (maintenance, repair, and operations) supplies, and office essentials. The platform serves enterprise buyers across Southeast Asia — factories, facilities teams, engineering contractors — aggregating a large and fast-changing supplier catalog with cross-border fulfillment.
Industrial procurement is structurally different from consumer commerce. Buyers arrive with a part number, a specification, or a photo of a worn component, and they need the right item — or a compliant equivalent — with certainty about fit, lead time, and regulatory constraints. That makes the catalog, not the checkout, the heart of the product; and EEZEE's catalog grows every week as new suppliers onboard with documents that were never designed for search.
The Challenge
An unstructured, long-tail supplier catalog
Millions of SKUs arrive as supplier PDFs, spreadsheets, and spec sheets with inconsistent naming, units, and categorization. Search quality degraded as the catalog grew, and buyers who could not find an item assumed it did not exist — lost revenue hiding inside messy data.
Slow, manual RFQ processing
Enterprise requests for quotation often run to hundreds of line items. Each list was parsed by hand, matched against the catalog, priced, and checked — a process that typically took days, during which impatient buyers sent the same RFQ to competitors.
Cross-border complexity checked by hand
Tax codes, restricted-item rules, and logistics constraints vary across SEA markets. Compliance checks lived in spreadsheets and staff memory, so cross-border quotes carried real risk of error — and every manual check added hours to turnaround.
Support team drowning in specification questions
Buyers constantly asked whether part A was compatible with machine B, or whether a cheaper equivalent met the same spec. These questions required real expertise to answer and consumed a large share of the pre-sales team's day, slowing response times for everyone.
The PULSAR Solution
PULSAR deployed a procurement copilot on its RAG and workflow platform on AWS, embedded directly into EEZEE's marketplace. The copilot treats the entire supplier catalog as a knowledge base — searchable in natural language, checkable for compliance, and usable for automated quotation:
Catalog intelligence pipeline
Supplier documents are normalized into a unified, searchable taxonomy — extracting specifications, units, and equivalency relationships automatically with managed foundation models. The pipeline achieves over 95 percent automated normalization, with low-confidence items queued for quick human verification. Search quality now improves with every new supplier onboarded, instead of degrading.
RAG-powered shopping assistant
Buyers describe what they need in plain language — 'a 316L stainless ball valve, DN50, PN16, food-grade certified' — and the assistant retrieves matching and equivalent items with cited specifications. Compatibility and equivalency questions are answered instantly with references to the underlying spec sheets, and compliant alternatives are suggested when an exact item is unavailable.
Automated RFQ processing
Uploaded quotation lists are parsed, matched to catalog items, priced against current agreements, and returned as draft quotes for human approval. A 200-line RFQ that previously took about four days now comes back as a reviewed draft within roughly four hours — with every unmatched line item explicitly flagged rather than silently dropped.
Market-aware compliance checks
The quote workflow embeds per-market rules: restricted items, documentation requirements, and cross-border logistics constraints are checked automatically before orders are committed, with flagged items routed to the compliance team. Risky quotes are caught before they become costly mistakes.
Outcomes
RFQ turnaround is where buyers feel the difference most directly: requests that used to take days now return as accurate drafts within hours, and EEZEE wins deals that previously went to whoever answered first. Search-to-quote conversion rose 31 percent on assisted product discovery — evidence that buyers who can find the right part buy more.
Internally, pre-sales support tickets on specifications and compatibility fell by nearly half as the assistant absorbed routine questions, freeing the team for complex, high-value consultative work. The catalog pipeline now scales with supplier growth rather than fighting it.
4 days → 4 hours
Typical RFQ turnaround for 200+ line-item requests
+31%
Search-to-quote conversion on assisted product discovery
-48%
Pre-sales support tickets on specs and compatibility
95%+
Supplier catalog items normalized automatically into the unified taxonomy
What's Next
EEZEE and PULSAR are extending the copilot to the supplier side — automated onboarding and catalog self-correction — and piloting demand forecasting to pre-position fast-moving industrial items across regional fulfillment nodes.
“Procurement buyers don't want to browse a million SKUs — they want the right part, compliantly, fast. PULSAR's copilot made our catalog feel like a knowledgeable sales engineer instead of a spreadsheet. Our buyers noticed before we told them anything had changed.”
Ready to write your own case study?
Talk to our team about deploying the PULSAR enterprise AI platform in your organization. Full client references available under NDA.