We translate advances in artificial intelligence into tangible business outcomes.
AI is moving faster than most companies can absorb.The constant hype (especially on social media) makes it hard to distinguish real opportunities from expensive distractions, leading to a lot of wasted effort.
There's a world of difference between a cool demo and a reliable tool. A demo has to work once in perfect conditions; a tool has to work every time, handle real world input, fit into your existing workflow, and produce results you can actually measure. This is where most AI initiatives fail, in companies big and small.
Large-scale AI projects often become another "digital transformation" initiative—heavy on overhead, light on immediate results. We've seen that the most effective organisations take a different approach. They start with a surgical improvement on a well-understood business problem. This delivers value quickly, reduces risk, and builds the crucial in-house muscle for using these new tools effectively.
Connecting the dots between what is technically possible, what is valuable for a business, and what it takes to ship a great product is the entire challenge.
This is our approach.
The best way to understand our approach is to see it in action. Here are some applications and engagements we've developed from our research.
Problem: Crafting a high-quality slide presentation is harder than it looks, even for the best managers. It requires clear thinking, a knack for storytelling, a way with words, and of course, an eye for design. Very few people have the training, or the time, to do this well. Most "AI slides" tools focus on template filling, ignoring the storyline and high quality visuals.
Improvement: We are building a set of AI agents trained on the slide-making principles we learnt from top-tier strategy consulting. Our system assists a user from a prose narrative to a logically-structured, client-ready presentation, ensuring the story is as sound as the slides are clean. It also provides MBB-level analysis and feedback on the user's own slides, to make them better - both visually and for communication impact.
Status: Currently in private beta with a select group of organisations.
Problem: A leading strategy advisory firm was looking to augment their consultants' expertise by unlocking years of proprietary knowledge and academic rigor stored across disconnected systems. They were looking to build an AI advisor, but without sacrificing the quality of insights or the user experience.
Improvement: In a long-term partnership, we've designed and built the core agentic architecture for their knowledge management systems. We're solving complex challenges in using AI to advise on strategy, collaborating with a university research lab, and building for agent reliability in high-stakes production environments.
Status: An ongoing, private engagement.
Problem: When we first started thinking about AI in sheets/excel, spreadsheets were still a manual frontier for AI. We saw an opportunity to bring agent-like capabilities directly into this universal tool, allowing users to easily chain calls and use their favourite LLM in a well known sheets environment.
Improvement: We built one of the first flexible and simple tools to bring large language models directly into Google Sheets. Having benefited so much from the open-source community, we gave the project back.
Status: Now an open-source project with growing use and thank you emails to the team. In a few cases, we have helped organisations build bespoke, proprietary extensions to this tool for their specific needs.
Problem: Technical recruiting cycles are slow, and screening engineering talent is a major bottleneck.
Improvement: We developed a suite of sophisticated ranking and filtering agents capable of performing expert-level candidate analysis, significantly reducing time-to-screen for our pilot partners.
Status: Following a successful pilot, this research project is now concluded. We have incorporated the agent-eval and orchestration modules into our ongoing work.
Ashwin has a strong track record as a product leader, from leading large teams at Waymo and Google to advising executive teams with McKinsey & Company. He also holds several technical patents. He is responsible for the lab’s direction, focusing its deep technical capabilities on creating product and commercial success.
We selectively partner with leadership teams to apply our expertise and focused approach for their most critical business objectives.
If you are working on a compelling problem, we invite you to reach out to us at projects@widgetlabs.ai.
Occasional essays on the practice of building reliable AI systems.
Why every LLM application needs a tailored knowledge retriever. We compare off-the-shelf RAG tools with a custom-built knowledge graph, demonstrating 4.2x better performance on relationship extraction.