Jagdamba Pratap Bhatt
I build the parts of AI everyone else skips.
Founder of HindustaniAI. I started as an aircraft maintenance engineer, spent a decade inside real businesses, and have built with AI since ChatGPT 3.5 — shipping a full AI platform and 250+ inventions in the hard parts: memory, honesty, safety and automation.
Built, not talked about.
Figures are from my own source-control and filesystem audit — self-documented and evidence-linked, not peer-reviewed or patent-granted.
From aircraft systems to AI systems.
I began as an aircraft maintenance engineer — an environment where precision isn’t optional and a single overlooked failure can have consequences far beyond the original mistake. That taught me how complex systems are really built: process, testing, documentation, verification, failure-prevention. Then I spent the next decade as a founder, working across many businesses — building their training videos, explainer videos, workflow and SOP systems — and learning first-hand how companies actually run: where leads leak, where work repeats, where information gets stuck, where approvals stall.
I’ve worked with AI since ChatGPT 3.5 and never stopped. I didn’t treat it as a chatbot — I treated it as a new layer for building reliable systems. I built Shrinam end to end, plus 100+ specialised tools, multiple B2B products and GTM automations. And I went deep on the problem most people overlook: an AI that can’t remember what happened, can’t explain why a decision was made, and can’t tell what it knows from what it’s guessing isn’t really operating — it’s just responding. So I built the memory, honesty and safety layers that make it actually work.
“In a plane, a tiny mistake can cost lives. In a business, a careless automation can cost crores. I build like it matters — because it does.”
Real products, running in production.
A multi-tenant AI platform where any business spins up its own voice-and-chat AI employees — AI SDR, AI Teacher/Trainer, Executive Assistant, outreach and embeddable website agents — with a talking avatar, long-term memory and per-tenant billing. Live in production.
An India-first AI-automation & GTM-engineering training platform (this site) — training the AI Operators and Pilots who run automated businesses.
A memory-as-an-API service that stores over a million pieces of history, answers only from what is genuinely saved, and issues a signed proof-of-erasure when data is deleted.
A marketplace where anyone can stand up a permanent public AI employee with its own phone-number handle — it lists its services, quotes a price, does the job, and earns.
An autonomous worker that actually does digital tasks — browses, fills forms, runs code — with an aviation-style per-step checklist and a human sign-off on risky actions.
Human-approved LinkedIn, email and WhatsApp prospecting that sources, scores, researches, writes, sends, replies and books meetings.
A self-hosted, drop-in replacement for a paid third-party messaging API — removed the subscription cost while keeping the exact same response format.
A persistent multi-store memory with knowledge-graph extraction, cause-and-effect inference, and idle-time “dream” synthesis — the shared brain behind the products.
A large server (200+ modules) that exposes the memory, governance, voice and reasoning engines as callable tools for any AI assistant.
A hundred-plus smaller tools built for specific business use-cases, plus multiple B2B products and GTM-engineering automations.
The hard problems — in plain English.
Over 250 inventions and architectures, many of which I found no existing prior art for. Each has a formal name and spec — but here is what they actually do, without the jargon. This is the engineering most AI products don’t have.
AI memory that answers only from what it genuinely knows — and says “I don’t have that” instead of inventing an answer. Delete something and it gives signed proof the data is truly gone.
Remembers not just what happened, but why — how one event led to another — so it can explain the reasoning behind a decision, not just repeat facts.
A lightning-fast working memory that surfaces what matters most to each person, ranked by how important it is to them — so it picks up exactly where it left off.
A real third option beyond “do it” or “fail”: the AI can pause and say “I’m not sure — let me check or ask” instead of confidently guessing wrong. This one idea stops it before a confident mistake.
A safety layer wired into the core that checks every action and either allows it, pauses it for a human, or blocks it outright — and it can’t be talked around or trained away.
Decides when to speak up on its own — because there’s a genuine reason, not a timer — so the assistant feels naturally attentive instead of silent or spammy.
Can genuinely erase one person’s data on request and hand you a signed certificate that it’s gone — built for privacy law and real trust.
Recognises who’s speaking from the sound of their voice alone — no login, no password — so a returning customer is known the moment they speak.
Knows the difference between “I never knew this” and “I knew it but it was deleted” — so it never quietly fills a hole with a made-up answer.
Reads the true drive and mood behind what someone says — tells genuine anger from mere excitement — and responds in the right tone instead of a canned reply.
Learns from whether things actually worked out — leans toward what succeeded, stays careful with what failed — so it improves on its own instead of repeating mistakes.
Every sensitive action — a consent, a deletion — is cryptographically sealed so no record can be quietly altered later. The signatures are built to stay secure even against future quantum computers.
Give it a company’s website and it maps the whole business — departments, roles, the step-by-step work each does — and pinpoints exactly where AI saves the most time and money, with the ROI.
Before it speaks, it checks its own answer — did I invent a name or a number? does it contradict what I said? — and holds back anything that looks made up.
Things I’ve actually done.
Built a US go-to-market engine from zero to a sustained 30+ qualified meetings a month through LinkedIn alone, within three months.
Personally closed $63,000+ in revenue across roles; consistently hit 220% of sales targets in a prior career.
Scored 94.1% strict / 96.0% fair with zero hallucinations on the LongMemEval long-term-memory benchmark; the honest-memory engine caught all 15 planted false-fact traps and correctly refused every question whose answer was not stored.
Produced a 6-minute AI short film before Sora existed and before image models had character consistency — solving consistency through disciplined, shot-by-shot prompting.
Recovered a 46,178-prompt / 345,638-reply corpus of real AI usage from fragmented, “corrupt” archives that every standard tool rejected — then instrumented it to measure where AI assistants silently fail.
Replaced a paid third-party messaging API with a self-hosted, contract-compatible service — eliminating the subscription while keeping the response format identical.
Shipped 3 products to production across 3 domains as a solo engineer, every product bilingual Hindi/English.
Career arc: Aircraft Maintenance Engineer → animator & creative lead → AI founder — with brand work for Redbus, Kores India, Ventura Securities and MI Lifestyle, and 11,000+ LinkedIn followers.