6+
Years active
40+
Engagements
4.8
Average rating
94%
On-time delivery
Client Reviews
In their own words
Ravindran Krishnan
Head of Digital Operations · Raffles Place, SG
"We had been running a customer query classifier for about eighteen months before engaging Thoughtium for the NLU Audit. The diagnostic report identified four specific failure modes we had not been measuring — including a pattern of misclassification on bilingual inputs that was affecting a significant portion of our Singapore users. The improvement paths were ranked clearly by impact. We acted on the top two within a month."
NLU Audit · February 2026
Lim Mei Ling
Content Director · Marina Bay, SG
"The Content Pipeline engagement was more structured than I expected. The two-week pilot phase was genuinely useful — it surfaced a mismatch between our brand voice guidelines and how we had described them internally. Arjun asked questions that made us think carefully about what we actually wanted the system to produce. The final pipeline has been running for three months without requiring Thoughtium's involvement, which was the point."
Content Pipeline · December 2025
Tan Hui Ning
Chief Technology Officer · Tanjong Pagar, SG
"We are a healthcare data company, and our ability to collaborate on model training is constrained by what MoH and our partners will permit. Siti understood the constraints immediately — she had worked in a similar environment before. The architecture she designed is genuinely privacy-preserving, not just marketed that way. The technical blueprint gave us something we could take to our legal and compliance teams and have reviewed properly."
Federated Architecture · January 2026
Priya Muthukrishnan
VP Strategy · One Raffles Quay, SG
"I appreciated that Wei Chen was direct about what the NLU Audit could and could not tell us. He did not oversell the scope. The report was twenty-three pages and every page was relevant — no padding, no generic AI commentary. What I particularly valued was the way findings were framed: not as problems to be solved by purchasing more services, but as a map we could use ourselves."
NLU Audit · January 2026
Ahmad Fauzi Jaafar
Head of AI Initiatives · Clarke Quay, SG
"The timeline for the Content Pipeline extended slightly during week four when we asked to add a second content type mid-engagement. Thoughtium was upfront about the scope change, sent an updated agreement, and completed the expanded work within the original end date. That kind of communication is not common. The pipeline itself has been working well — our editorial team uses it daily."
Content Pipeline · November 2025
Yong Chee Kiong
Data Science Manager · Shenton Way, SG
"We had a complex multi-node scenario involving three separate subsidiaries with different data governance requirements. The Federated Learning Architecture engagement had to reconcile constraints from each entity. Siti navigated that with patience — she spent more time on the scoping than we expected, but by the time the architecture work began, there were no surprises. The reference implementation worked on the first joint test."
Federated Architecture · December 2025
Case Studies
Engagements in detail
Challenge
A mid-size wealth management firm had deployed a chatbot for client queries eighteen months prior. Client satisfaction scores for chatbot interactions were consistently lower than for human interactions, but the team could not identify why — internal testing had not revealed obvious problems.
Solution
The NLU Audit ran structured tests across 340 input scenarios, including code-switching inputs common in Singapore (English/Mandarin, English/Malay). Testing identified three failure mode clusters: intent ambiguity at conversation boundaries, mishandling of negation patterns, and near-zero accuracy on numeric queries involving local financial terminology.
Results
The diagnostic report provided ranked improvement paths. Six weeks after report delivery, the firm's internal team had addressed the top three recommendations — improving intent accuracy by 31% on the identified problem categories. The reusable test framework is now part of their quarterly review process.
Challenge
A professional services firm needed to produce substantially more content — case study summaries, service descriptions, and newsletter articles — without increasing the editorial headcount. Initial experiments with off-the-shelf AI writing tools produced output that consistently failed to match the firm's established voice and professional standards.
Solution
The six-week Content Pipeline engagement began with a detailed analysis of the firm's existing editorial archive to derive voice characteristics that had not been explicitly documented. Prompt engineering was built around these characteristics, with quality controls calibrated to the firm's editorial standards. Human review workflows were mapped to the firm's existing approval processes.
Results
The pipeline now produces first drafts that the editorial team describes as requiring "light editing rather than rewriting." Content production volume increased without additional headcount. Three months post-engagement, the pipeline is operating without Thoughtium's involvement, as designed.
Challenge
A healthcare data company and two partner institutions wanted to jointly improve a clinical text classification model but faced strict constraints — patient data could not leave its originating environment under any circumstances, and each institution had different technical infrastructure.
Solution
The twelve-week engagement designed a federated architecture with differential privacy mechanisms appropriate to the clinical context. Communication protocols were designed around the lowest common denominator of the three institutions' infrastructure. Security review included PDPA and MOH data governance requirements. A reference implementation was built and tested across all three environments.
Results
All three institutions were able to participate in joint model training without any patient data leaving its source environment. The model's classification accuracy on the combined task improved compared to models trained on individual datasets. The technical blueprint was subsequently reviewed and approved by each institution's compliance team.
Credentials
Professional standing
Registered in Singapore
ACRA-registered business entity · Singapore 049315
AISG 100 Experiments Mentor
AI Singapore · 2023
PDPA-compliant practice
All engagements conducted under Singapore PDPA
Add your experience to this list
Begin with a conversation. We respond to all enquiries within one business day.
Contact Thoughtium