Berlin · Co-founder and CTO, Bryo · LLM experiments
Nikhil Mudholkar
Co-founder and CTO of Bryo in Berlin. We build technical sales AI for manufacturing companies. I also run independent experiments on how LLMs and decision models behave in real work.
Independent experiments · 2026
Experiments
A decision model against two Gemini models. Jev is a little less accurate, costs a tenth as much, and is the only one whose confidence you can trust.
Ask the what-ifs before you choose. An 8-step agent loop drops to 3 calls and gets more accurate on both Jev and Gemini.
Jev, Perplexity and Cloudflare Clef side by side. Perplexity wins overall. Clef is accurate but underconfident and slower than claimed.
Diogo Almeida, CEO of TypeSafe, the company that makes Jev, quoted the thread:
yay for private evals and calibrated confidence scores! 🥹 go automation!
5/8 For email routing, the question is: Which emails can we route automatically, and which need a person? Jev’s most confident 85.5% had zero errors against the reference labels. That’s on this dataset.
Work
Experience
- Led the onboarding team. We turn supplier catalogues and ERP data (bilingual PDFs, technical drawings, millions of product variants) into a structured knowledge graph.
- Defined the product ontology across 1,000+ categories and tens of thousands of specifications. Automated QA of that data, first with freelancers, then with LLM-as-judge pipelines.
- Built a multi-step search pipeline that reasons over the ontology, then led the team through the move to agentic search with custom harnesses for our industry.
- Built integrations with SAP ECC, Odoo, Business Central, Outlook and SharePoint.
- Own infrastructure, uptime and LLM infrastructure: EU-region model hosting, LiteLLM, Vertex AI, Langfuse. Built Bryo's framework to benchmark LLMs and agents across the stack.
- Responsible for GDPR compliance and ISO 27001 (in progress). Led security audits with enterprise customers.
- Raised about $3M from investors including Entrepreneur First, HTGF and Playfair Capital, plus about $0.6M in grants and $0.6M in loans from the Berlin Senate and IBB.
- Led technical pre-sales. Hired and managed 8 engineers and 2 commercial staff. Scaled the product to customers in production with six-figure revenue.
- One of five engineers running the data infrastructure. Built pipelines that served about 30 data analysts. Python, Airflow, AWS, dbt.
- Built security tools for Snowflake, S3, RDS and Oracle from CloudTrail and query logs. Used user and entity behaviour analysis to flag possible data breaches for insider threat teams.
- Built a graph-database tool to monitor critical processes and limit the impact of failures.
- Received the Phoenix Data Award, a company-wide award for major data breakthroughs.
- Internship, summer 2018: built a SAS-to-Python transpiler with ANTLR4 that converted 11,000+ SAS scripts to PySpark. Received a pre-placement offer.
Education
IIT Bombay
- KVPY fellowship, given each year by the Government of India to about 2,000 of 100,000 students for research.
Outside work
Side projects
- Automated derivatives trading. Backtested strategies on five years of market data and built a system that trades intraday from live broker ticks. Strategies include hedged theta decay on weekly index options and pair trading on co-integrated stocks.
- Technical analysis screener. A web app that scores stocks in a portfolio and gives trade recommendations.