Skip to content

Systems onlineMumbai · UTC+5:30

Hitansh Gopani — AI/ML EngineerI build AI systems that keep working after the demo is over.

Turning emerging technology into systems, products and workflows that solve real problems — as an AI/ML Engineer at Pitch Perfekt Collective and co-founder of LightNoteAI.

  • AI systems
  • LLM evals
  • Agents
  • GenAI
  • Automation
Explore work →Download CVGet in touch

eval · hitansh

  1. Ships working softwareLive demos: Crucible, VoxGate, Sigmanot run
  2. Evaluates before trustingTrustLLM prototype · honest baselines in Incident Intelnot run
  3. Documents what broke14 findings logged in Cruciblenot run
  4. Leads & builds teamsCo-founder · 15+ devs mentorednot run
  5. Explains clearlyTrained professionals & students · Medium articlesnot run
Building AI systems since Nov 2023
Building AI systems since Nov 2023
Production builds with public code
Production builds with public code
Automated tests in Crucible
Automated tests in Crucible (opens in a new tab)
Professionals and students trained
Professionals and students trained
Colleges taught at
Colleges taught at
Developers mentored
Developers mentored

Explore as

The 30-second version

  • 2+ years building AI systems (since Nov 2023) — now AI/ML Engineer at Pitch Perfekt Collective (Jul 2026–) · Co-founder, LightNoteAI (Oct 2025–)
  • Prior: AI/ML Trainer at QUASTECH — 50+ trained, sessions delivered at 5+ colleges; Radical AI (AI software, remote); Vendiman (data)
  • Strengths: LLM systems & evaluation, agents, voice, vision ML, shipping to production
  • Open to conversations · Mumbai, UTC+5:30 · remote-friendly

Benchmarks

Selected work

Each project: the problem, the demo version, and what it took to reach production — plus what broke, where I've written it up.
  • In progressLLM Systems & Evals

    TrustLLM

    Test your LLMs before your users do.

    ProblemTeams ship LLM features without knowing how they fail — hallucinations, drift, unsafe output, unknown cost.

    Prototype · up to 4 models · 5 metrics + composite (simulated outputs)

    DemonstratesLLM evaluation · Guardrail design · Product thinking

    Case study →
  • LiveVerifiable AI

    Crucible

    Every fine-tuned model gets a birth certificate.

    ProblemWhen you fine-tune on someone else's GPU, what can you prove about where the model came from?

    1,327 tests · 14 documented findings

    DemonstratesSystems engineering · Testing discipline · Debugging under pressure

    Case study →
  • LiveVoice & Agents

    VoxGate

    Voice interviews that gate compliance decisions — explainably.

    ProblemOnboarding compliance (sanctions, PEP, adverse media) is slow, manual and hard to audit.

    Pack architecture · KYC-UAE launch pack

    DemonstratesAgent design · Voice interfaces · Regulated domains

    Case study →
  • ShippedVision & ML

    Incident Intel

    Spot what's unusual in fixed-camera footage — and say how sure you are.

    ProblemSecurity and production-line cameras record everything; people only need the unusual moments.

    3 approaches compared · baseline AUC 0.709, deployed autoencoder 0.674

    DemonstratesML rigour · Data versioning · Honest metrics

    Case study →

Also in the lab

I like building things that shouldn't exist yet.

Some started as experiments. Some became products. Some broke spectacularly. All of them taught me something.

01 · Why I build

Building AI isn't the hard part.Keeping it working is.From data,to intelligence,to production.

Hitansh Gopani in a black suit, arms crossed

Hitansh Gopani

AI/ML Engineer · Mumbai, India

About the model

Engineer. Builder. Thinker.

I'm an AI engineer in Mumbai. By day I build production AI pipelines at Pitch Perfekt Collective; I also co-founded LightNoteAI, the intelligence layer for proposals.

I care about the unglamorous part — evaluation, failure modes, tests and logs — because that's what decides whether an AI system is still trusted a month after launch.

I learn in public: 15+ developers mentored, articles on Medium, and a habit of writing down what broke.

Value

What I bring to the table

  • 01

    Technical depth

    Across the AI stack — data pipelines, classic ML, LLM apps, RAG, agents, APIs and deployment.

    4 end-to-end builds with public code · 1,327 tests in Crucible

  • 02

    Systems thinking

    I don't treat AI as an isolated model. I design how it connects to data, infrastructure, people and workflows.

    System map above

  • 03

    Builder energy

    I prototype, integrate and ship — and when a demo crashes 12 hours before a deadline, I rebuild it.

    Avishkar zonal win · Code Odyssey runner-up

  • 04

    Business orientation

    I don't only build inside existing systems — I find problems and turn them into products. The goal isn't clever technology; it's something useful.

    LightNoteAI · McKinsey.org Forward

  • 05

    Builds for other markets

    Compliance logic written against a foreign regulatory regime, and AI work delivered remotely to a team I never shared a room with.

    VoxGate KYC-UAE pack · Radical AI, remote

Intended use

The system I build

Every project I take on lives somewhere on this map, with the tools behind each part.

Trace a project

module · evals

Evals & guardrails

Hallucination, grounding, consistency and toxicity checks before anything reaches users.

  • Grounding
  • Hallucination
  • Toxicity
  • Audit logs

Proven in

Out of scope · Demos that only need to work once.

Leadership · Co-founder · Oct 2025 – present

I co-founded LightNoteAI to help businesses win more proposals.

Co-founded in Oct 2025, during the final year of my B.Tech. The intelligence layer for proposals.

  1. The problem

    Proposals reach clients without anyone checking their quality, their weak spots, or what the buyer will actually evaluate.

  2. What we're building

    An AI proposal-intelligence platform that analyses proposal quality, finds weaknesses, surfaces buyer and evaluation signals, and recommends fixes — before the proposal goes out.

  3. What I work across

    • Product
    • AI
    • Strategy
    • Growth

    Focus areas

    • Proposal scoring
    • Win intelligence
    • LLM workflows
    • SaaS infrastructure
  4. Visit lightnoteai.com (opens in a new tab)

What co-founding proves

  • 01

    Initiative

    I didn't wait for a job title to build a product. I spotted a problem and co-founded a company around it while still a student.

  • 02

    Leadership & ownership

    I work across product, AI, strategy and growth — the decisions, not just the tickets.

  • 03

    Real problem-solving

    Proposal quality is a business problem with money on the line. The AI has to change an outcome, not just produce text.

  • 04

    Commercial thinking

    Product strategy and go-to-market sit next to the model work. Building something businesses will adopt is its own discipline.

Training data

Experience

Everything I've worked on, newest first — with co-founding LightNoteAI highlighted as leadership. B.Tech in Information Technology, K. J. Somaiya Institute of Technology (2022–2026) — CGPA 8.67/10, with Honours in Blockchain & Distributed Ledger Technology.

  1. Jul 2026 – present

    Artificial Intelligence Engineer · Pitch Perfekt Collective

    • Production AI pipelines that keep creative assets, characters and workflows organised, reusable and ready for production.

  2. Dec 2025 – Feb 2026

    Training · Public speaking

    AI/ML Trainer · QUASTECH

    • Trained 50+ working professionals and final-year students in AI and ML — including career-changers moving in from other domains, for real projects and placement readiness.
    • Ran QUASTECH's internal AI sessions and workshops, and delivered end-to-end sessions and training at 5+ colleges.
    • Hosted and led sessions, workshops and seminars throughout — live technical delivery and public speaking were a core part of the role.
    • Designed and delivered the curriculum on ML, transformers and GenAI; mentored learners through their own projects.
    • Built the course's end-to-end RAG pipelines: chunking, embeddings, FAISS indexing, MMR retrieval and LangChain retrieval chains, served through FastAPI and Docker.
    • Kept content current as AI tools and industry use cases changed month to month.

  3. Oct 2025 – present

    Leadership

    Co-Founder · LightNoteAI

    The intelligence layer for proposals. I work across product, AI, strategy and growth.

    What co-founding it proves →

  4. 2024 – 2025

    Mentorship

    Facilitator · Google Cloud Arcade

    • Co-ran a mentorship community for Google Cloud Arcade learners.

  5. Jun – Jul 2024

    Contributor · GirlScript Summer of Code

    • Open-source contributions during GSSoC'24.

  6. May – Sep 2024

    Data Analyst · Vendiman

    • Automated SQL reporting and dashboards in Metabase — about 60% less manual reporting effort.
    • Built WhatsApp automation workflows.

  7. Nov 2023 – Apr 2024

    AI Software Developer · Radical AI

    • Built AI-driven education apps in Python and web tech; open-source work on access to AI.

  8. Sep 2023 – Jul 2024

    Team leadership

    Technical Manager · IoT Cell, KJSIT

    • Led a 15-member engineering team across architecture, code reviews and debugging.
    • Led the build of a full e-commerce platform with Stripe payments.
    • Mentored 15+ developers on building and deploying full-stack apps.

  9. Jun – Sep 2022

    Outreach Coordinator · Google Developer Student Clubs, KJSIT

    • Outreach for community events including “Sprint”.

In the room

  • Delivering a session — QUASTECH, Mumbai

Writing

Worked out in public

Articles, talks and workshops, with links.

Talks & workshops

Principles

What I care about

Five working principles, each with the thing that taught it to me.

  • 01

    Dependable over impressive

    “Building AI models isn't the hard part. Building AI systems that people can rely on every day is.”

  • 02

    Own it when it breaks

    Our Avishkar demo crashed 12 hours before the deadline. We rebuilt it on three hours of sleep — and won the zone.

  • 03

    Teach to understand

    Explaining transformers to beginners showed me the cracks in my own understanding.

  • 04

    Build for access

    At Code Odyssey we built for people who dream in handwritten notes and their own languages.

  • 05

    Stay in rooms where you're not the smartest

    I'm often the youngest in the room. It makes me listen differently — and stay anyway.

Changelog

Build log

  1. v3.0In training.
  2. v2.3Sep 2026Dell Technologies Forum, Mumbai — notes on moving AI from impressive to dependable.
  3. v2.2Jul 2026AI/ML Engineer at Pitch Perfekt Collective. CSCMP Generative AI for Supply Chain certificate.
  4. v2.1May 2026Graduated — B.Tech IT, K. J. Somaiya Institute of Technology.
  5. v2.0Mar 2026Snowflake “Zero to Agents” — guided hands-on workshop: a governed agent with text-to-SQL and RAG.
  6. v1.9Dec 2025AI/ML Trainer at QUASTECH · Zonal winner at Avishkar · McKinsey.org Forward.
  7. v1.8Oct 2025Co-founded LightNoteAI.
  8. v1.7Sep 20251st Runner-Up, Code Odyssey 4.0 (Tata STRIVE × KJSIT) — SapnaForge.
  9. v1.52024 – 2025Google Cloud Arcade facilitator · workshops and mentoring.
  10. v1.02023 – 2024First AI internship (Radical AI, remote); led IoT Cell; Vendiman data internship; Postman Student Expert.
  11. v0.12022Started at KJSIT; GDSC outreach.

Awards & certifications

Research interests

Questions I'm exploring

Ask Hitansh · the intelligence layer

Ask about my work

Every node below is real — a skill, strength, project, role or research question from this site, linked the way the work actually connects.

Retrieval over this site's verified content · runs in your browser · no LLM, nothing leaves the page.

Console

Prefer a terminal?

The whole site, as a shell. Try whoami, ls projects or open crucible.
hitansh.os — zsh
hitansh.os — type help, or tap a command below.
hitansh@lab:~$ whoami
Hitansh Gopani — AI/ML EngineerI build AI systems that keep working after the demo is over.Mumbai, India · open to conversations

How to reach this model

Let's build something that survives production.

Email me →GitHub(opens in a new tab)LinkedIn(opens in a new tab)Medium(opens in a new tab)X(opens in a new tab)

Mumbai · UTC+5:30 — working hours overlap European mornings and US evenings.