Bhavya Sharma
Trainee Associate @ Freshfields
Building technical products that give lawyers real value
About
I’m a trainee solicitor at Freshfields, currently on secondment in Silicon Valley. I was previously the President of the legal tech society at King's College London and have interned at a few legal tech start ups. I’m interested in the point where legal expertise stops being trapped in documents, emails and individual memory, and starts becoming structured, reusable infrastructure. I’ve been building in that direction. My main project (private repo) is an AI-native legal workspace for matter teams, designed around context preservation, document/version workflows, correspondence filing, trackers and matter-level knowledge. I recently put out Cr_oss (open-sourced), an Outlook add-in that lets you track changes natively, and am working with the Freshfields AI Lab to have this rolled out firm-wide. I’m also building CompMap (open-sourced), a source-first competition law research tool that maps market definition reasoning across EU, UK and US cases, with propositions grounded back to primary materials rather than retrofitted citations (which has proved very helpful in avoiding hallucinations). I’m not drawn to legal tech because it makes existing workflows slightly faster. I’m drawn to it because lawyers understand the hidden logic of legal work better than anyone else, and should be building the systems that reshape it. I’m still early, but I’m shipping, learning quickly, and building the tools I wish existed in practice.
2 Projects
cr_oss
Privacy-first Outlook compose add-in that brings Word-style track changes to email drafts. No backend, no login, no analytics - everything runs locally in the task pane.
Meridian
Open-source market-definition research graph and merger control analysis tool for competition lawyers. Built a pipeline that combines deterministic checks and LLM calls to extract case content. Currently at ~250 cases fully through the pipeline (have spent around $30 in API credits, while using the Gemini AI Studio free tier as much as possible). Built a database, scraped from official authority sources, that contains merger control thresholds and rules for the biggest 30 jurisdictions (and growing). Created an intake form that allows a user to feed information it has, and created a screening result based on the official sources. Very much maps my workflows as a competition lawyer. CompMap lets you search merger precedent across the EU, UK, and US by sector, product market, authority, theory of harm, and outcome. Every market-definition proposition links back to a specific page, paragraph, and quote in the underlying decision or court document (source-first philosophy).
Philosophy
"Lawyers should not just adopt tools, they should understand enough to reshape the workflow."
The real leverage is not in making old processes slightly faster. It is in lawyers using their domain judgment to redesign how legal work is captured, verified and delivered.
"Data-first, source-first, always."
Legal AI is only useful if its outputs can be traced back to real materials. I care about systems where propositions and generations are grounded in source text / real data from the start, not dressed up with citations afterwards.
"The best legal tools come from people close enough to the work to feel the friction."
A lot of valuable legal software starts with small, specific pain points that practitioners understand intimately. I’m interested in building from those pain points outward: shipping useful tools, testing them in real workflows, and only then deciding what should scale.