Writing & appearances

1. 24 OctAI as an investigator, not just a builderThis week, PHP released fixes for two critical vulnerabilities. The underlying issue had been suspected as early as 2024, but no one could show a concrete attack. Later AI-assisted research helped demonstrate how it could allow an attacker to execute code remotely, making the full impact clear.LinkedIn (opens in a new tab)
2. 15 OctBuild AI capability inside the existing teamSometimes the skills needed to build a product can change faster than we can organise the team building it. For engineering leaders, that creates a recurring decision: when the work demands a new capability, do we build a separate team around it, or develop it within the existing team?LinkedIn (opens in a new tab)
3. 7 OctDecision models make workflow gates affordableMost agentic workflows run every step through a generative model. Even the steps that aren't generating anything. If we look closely, many steps are really decisions. Is this true or false? Which of these five categories does it fall in? How relevant is it, on a scale of one to five?LinkedIn (opens in a new tab)
4. 25 SepLeave capacity uncommitted for innovationOne principle I generally push for in engineering planning is to leave some capacity uncommitted and leave room for discovery, experimentation, engineering growth. An engineering org that wants to be driven by innovation needs some capacity that is not already committed to a known deliverable.LinkedIn (opens in a new tab)
5. 19 SepAI changes the economics of legacy migrationsA lot of legacy systems stay in production because the cost of moving away from them is difficult to justify. The team knows where the limitations are, and often knows what a better stack would look like.LinkedIn (opens in a new tab)
6. 8 SepHow our cloud bill kept falling while usage grewAt Elucidata our cloud bill has been going down for several quarters now, while usage and the product portfolio have been growing. When people ask what we did, they expect a technical answer, but most of it wasn't.LinkedIn (opens in a new tab)
7. 3 SepLet AI agents control what they reveal, not what they readShould we let AI Agents censor their own output instead of applying access control filters at source? Idea is uncomfortable, but I think we will end up there anyway. Most companies now have an AI agent over their internal knowledge base, answering questions from whatever has been indexed into it.LinkedIn (opens in a new tab)
8. 31 AugSending AI-generated text unedited is a cost transferA very common use of AI now is generating explanations, summaries, design reasoning, brainstorming output. And by default, most reasoning models return several paragraphs even when the actual answer is a line or two. The model doesn't know what you already know.LinkedIn (opens in a new tab)
9. 18 AugVersioned knowledge graphs: why an agent's shortlist changesAsk an agent to shortlist targets for an indication today. Ask again in 6 months and two names have dropped off. Nothing about the question changed. What changed — the biology, the source databases, or the ingestion underneath? On most systems nobody can say.LinkedIn (opens in a new tab)
10. 23 JunMCP servers for PollyFor years, we've delivered AI-ready biomedical data that teams trust to drive discovery. Now we're handing them the controls — letting people build the exact workflows their science demands.LinkedIn (opens in a new tab)
11. 9 AprSecurity blind spots in enterprise RAGMost enterprise RAG deployments have security blind spots that should keep you up at night.LinkedIn (opens in a new tab)

1. 8 SepFederated Learning in Healthcare AILearn how Elucidata’s federated learning enables secure, compliant AI-ready data collaboration - faster insights, lower costs and better generalizability.Elucidata whitepaper (opens in a new tab)
2. 28 JanScaling High-Quality Data Processing: How to Achieve 4x Cost Reduction for Foundation ModelsThis presentation will demonstrate how Polly achieves 4x lower costs and 2.5x faster processing speeds. Gain insights into a head-to-head comparison of Polly and AWS on cost, speed, and ease of use.Elucidata webinar (opens in a new tab)

1. 12 AugSecurity & Compliance on Polly - Accelerating Drug Discovery SecurelyExplore how Polly ensures security and compliance for accelerating drug discovery with robust data management.Elucidata whitepaper (opens in a new tab)
2. 7 AugRole of MLOps in Biomedical ResearchDiscover how MLOps is helping advance biomedical research, enhance drug discovery, and improve healthcare outcomes.Elucidata article (opens in a new tab)
3. 1 JulWhat is the right size of a team?I have faced this question multiple times whenever there is a change in team or new team is being created. I am sure other engineering managers(at junior level or senior level) face this situation.Essay

1. 14 NovChanging User Account Password in UbuntuEssay · Ubuntu
2. 1 FebEmbedding Terminal in GeditEssay · Ubuntu

1. 3 MarAutomatically Mounting Partitions in UbuntuEssay · Ubuntu