Built a tool that catches the insurance clauses that actually get claims rejected.
my sister's health insurance claim got rejected a while back over a clause buried on page 14 that nobody actually reads before signing up. turns out this kind of thing is way more common than i realized, some fine print condition that's technically disclosed but never actually explained to you when you're buying.
that stuck with me enough that i ended up building something for it.
it's a tool where you upload your policy pdf, or just answer a few quick questions if you'd rather not upload anything, and it flags the clauses that typically cause claim rejections. explains everything in plain language instead of the jargon insurers usually bury things in. covers health, term and motor insurance right now.
also works in hindi, marathi and telugu, since a lot of people here genuinely don't read policy documents comfortably in english, and honestly that language gap is part of why so many people get blindsided by fine print in the first place.
no login, doesn't save your data. sticks to just telling you what's in the policy, not whether you should buy it, wanted this to stay strictly factual instead of drifting into advice territory.
still testing, still finding rough edges myself. would genuinely appreciate if some of you tried it against your own policy and told me what it missed or got wrong
Solving the language barrier for regional Indian languages (Hindi, Marathi, Telugu) is huge here. Legal and insurance terminology usually gets completely mangled in standard machine translation, so getting plain-language regional outputs right is a massive win. Curious—how are you handling messy PDF extractions like scanned tables or low-quality scanned policy docs without triggering hallucinations on critical policy conditions?
Great use case for AI in everyday life! The idea of catching problematic clauses before they cause issues is exactly the kind of proactive tool people need. Have you considered adding a browser extension version so users can check clauses in real-time while browsing insurance documents? That would make it even more accessible.
thanks, really appreciate that! browser extension is a great shout actually, hadn't thought about the real-time-while-browsing angle specifically. right now it's a standalone upload/questionnaire flow but that's a genuinely useful direction to think about for v3. adding it to the list. link if you want to try the current version: https://tc-simplified-v2.netlify.app/
The "no login, doesn't save your data" choice is doing a lot of work here — for a tool reading someone's actual insurance policy, that's the difference between people uploading the real document versus a sanitized decoy. Curious how you're handling flagging accuracy without any stored corpus to validate against — pure clause-pattern matching, or an anonymized rule library kept separate from user uploads?
yeah exactly, that's the whole point of the no-storage choice, wanted people to feel safe uploading the real thing. accuracy comes from a rule library built off known irdai patterns and common rejection clauses, not from any stored user corpus. so it's more clause-pattern matching against that library than anything learned from past uploads. kept those two things deliberately separate. link: https://tc-simplified-v2.netlify.app/
That's the right split — a fixed rule library beats anything trained on user data for this, since the library doesn't leak what any one user uploaded. Tried the tc-simplified link, makes sense.
The line "tells you what's in the policy, not whether you should buy it" is the smartest part of this, it keeps you out of the regulated-advice conversation entirely while still delivering the value people actually need. The Hindi, Marathi, and Telugu support is probably the bigger moat than the clause-detection itself, most insurtech tools stop at English and miss exactly the population most likely to get blindsided by fine print.
appreciate that, and honestly agree on the language part, that's been more work than the clause detection itself but feels like the actual point of doing this in india specifically. wanted people to be able to trust what they're reading regardless of which language they think in.
Really practical build with a genuine purpose behind it.
Thanks, in the world full of AI, only purpose can make you stand out.
The “strictly factual” boundary is interesting here because users may still make important decisions based on what the tool flags.
I’d be curious which matters more in testing so far: catching more clauses, or making sure users understand why a flagged clause matters.
right now it's leaning more toward making sure people understand why something matters rather than maximizing how many clauses get caught. reasoning is pretty practical honestly, a flag that says "risky" with no context is basically useless, people just get scared or ignore it. the explanation is what actually lets someone go back to the insurer and ask the right question, or decide it doesn't matter for their situation.
also ties into staying on the right side of what's actually allowed here, insurance advice is regulated in india, so the tool sticks to explaining what a clause means and why it's worth checking rather than telling anyone what to do. that boundary actually pushed me toward prioritizing clarity over sheer coverage, since a confusing flag with no real explanation isn't that different from just giving advice by implication.
still early days testing this though, so take that as where i'm leaning, not a settled answer yet. link if you want to see what it actually looks like: https://tc-simplified-v2.netlify.app/
That makes sense. I’d be interested in continuing the conversation beyond the thread — would you be open to sharing the best email to reach you on?