VAAS — Detecting Image Manipulation with AI
Introduces an attention-based anomaly scoring approach that combines global verification with patch-level localisation for interpretable forensic analysis.
Read paper →We research it. We engineer it. We ship it as an unfair advantage - for the consumers who live with it daily, and the enterprises that need to stay ahead of it.
We don't hand you a slide deck. We design, build, and deploy your AI systems with measurable outcomes.
Autonomous agents with tool use, memory, and modular orchestration. Built for reliability in production environments.
High-fidelity retrieval pipelines grounded in your proprietary data. Hybrid search, re-ranking, and evaluation built in.
Vision–language systems, video understanding, and cross-modal representations at scale.
CUDA kernels, distributed training, inference optimisation. Squeeze maximum performance from your hardware budget.
MLOps pipelines, model registries, observability stacks, and serving infrastructure that teams trust in production.
Technical due diligence, capability audits, and a clear roadmap, so you invest in AI that compounds.
We run a free 20-minute scoping call to map your AI opportunity and cut through the noise.
Schedule a Free Scoping Call →Production AI products across legal tech, sales, and customer operations.
Fine-tuned RAG system that turns raw Nginx log errors into actionable fixes — served via vLLM on Modal and delivered as a Docker Desktop extension.
Learn more →AI-powered system that detects and localises image tampering, delivering global integrity signals and pixel-level evidence.
Learn more →In one of our mission to providing unfair advantage to our users, we are building Ludo - a jobboard agent that helps job seekers manage, merge cvs, orchestrate, prep and solve interview questions using agentic AI.
Learn more →Our team combines deep research expertise with hands-on engineering experience. We've built tools pioneering the frontier of enabling AI on the web. We've also worked at different institutions championing the ideation and the deployment of AI systems in production.
Technical deep-dives, applied research, and production lessons from the team.
Introduces an attention-based anomaly scoring approach that combines global verification with patch-level localisation for interpretable forensic analysis.
Read paper →A case study in building a fine-tuned RAG system for DevOps log debugging — from dataset curation and QLoRA fine-tuning to a Docker Desktop extension.
Read case study →
A practical framework for training LLM agents under real-world tool constraints like latency, failures, and non-differentiability.
Read paper →
A structured playbook for taking ML from prototype to reliable production — training, serving, logging, and monitoring included.
Read article →
A systematic review of deep learning methods that infer location from imagery — bypassing missing metadata in digital forensics.
Read paper →Let's scope your project in 20 minutes. No slides, no fluff, just a direct conversation about what's possible.