Siddharth Pachpor

An AI agent that negotiates with suppliers on its own: it reads their emails and offers, calculates the total cost of every contract, and decides when to counter, accept or walk away.

Built at the Industrial Agents Hackathon by UVC Partners x Anthropic, Munich 2026, where it won first prize (gold medal). I led the strategy and the pitch and worked on the code with my teammate.

Training staff to follow a written procedure is easy. Checking that they actually did is not. FlowVerify takes the WHO order of blood draw and checks a training video against it, frame by frame.

Built at the Google DeepMind hackathon at Stanford University, USA, in 2026, where it reached the final.

A satirical app for the DACH business crowd. You describe your own LinkedIn or Xing profile, and Claude turns it into an "archive card" that ranks you in ten tongue-in-cheek categories, with a famous match.

Built as a fun side app for the Wallnuss community. Live, with paid unlock.

A pre-seed startup wanted to move product planning from manual handoffs to an agentic AI workflow. Program managers were spending hours turning product ideas into stories and tasks; the brief was a reusable workflow, not a chatbot.

A planner splits the goal into steps, a router sends each step to the right specialist agent, and every output is checked before it moves on.

How do you keep users once they have signed up? In a Slack case I mapped the activation moments, then followed weekly user cohorts to see where retention breaks and what churn does to customer value.

On the case data, about half of users were still active after eight weeks, the biggest drop came right after the first week, and monthly churn fell while lifetime value per cohort rose.

Settling an estate in Germany can take one to two years and cost several percent of its value. Wealth managers' minimums leave most families out.

InheritIQ combines guided document creation, a digital register of assets and short-term loans against inherited assets, with notaries as partners for legal validity.

On its own, Tiffany was worth about what the market paid: $93.80 per share by DCF against a $91.58 share price, and $93 to $102 on peer multiples.

The case for the deal sits in synergies, about $9.0bn, of which about $7.8bn are cost and working-capital synergies that do not depend on selling more. Sharing 70% with Tiffany's shareholders gives a fair price of $145.21. At $135, LVMH keeps about $10 per share.

From 2016 to 2026, US stocks and corporate bonds moved together more than textbooks assume (correlation +0.57). That shrinks the benefit of mixing them.

On this data the best risk-adjusted portfolio was 100% stocks. Bonds alone earned less than cash.

Daily-wage workers in India's villages had little access to clear COVID-19 safety guidance. I built Project Vir, a virtual reality app for low-cost Google Cardboard headsets, to teach it hands-on.

Users see the virus up close, then pick a mask, gloves and sanitiser from a Hindi and English menu and place them in the room themselves. The Embassy of India in Berlin forwarded the app for review in 2020.

An office-supplies company was giving away margin on discounts and seeing too many products sent back. Our team modelled its order data in R to find what really drives profit, profitability and returns.

Discount explained most of the swing in profitability, and orders above about 30% discount were very likely to lose money. Returns were far more likely in the West region and with same-day shipping.