personalyear-in-reviewmlstartups

2024: IIT Roorkee, Writing, and the Funding Winter

3 min readUtso Sarkar

2024 was the year I joined IIT Roorkee in August and went deep on technical foundations while the Indian startup ecosystem contracted around me. This is the honest review: coursework, side projects, and the funding winter that forced clarity — without pretending I was running a venture-backed company.

January-July: Pre-IIT and Strato Inc

Before August, I was still publishing through Strato Inc, sitting board exams, and navigating JEE outcomes. No incorporated companies. Just maintenance, learning, and the hallway between school and college.

August-December: IIT Roorkee Begins

Joined B.S./M.S. Mathematics and Computing at IIT Roorkee in August 2024. Q3 and Q4 were coursework shock, automata theory, and learning to balance hostel life with the Play Store account I refused to abandon.

flowchart LR
    subgraph Research["Learning - 2024 Status"]
        R1[RAG Production Patterns - Documented]
        R2[Continual Learning Reading - Ongoing]
        R3[ViT vs CNN Deployment - Applied]
        R4[Multimodal LLM Skepticism - Written]
    end

    subgraph Skills["Skills - 2024 Status"]
        S1[IIT Mathematics - On Track]
        S2[ML Agent Reading - Side Interest]
    end

    subgraph Writing["Writing - 2024 Status"]
        W1[Technical Essays - 12+ posts]
        W2[Founder Opinion - India-specific]
        W3[Personal IIT transition posts]
    end

    Research --> Writing
    Skills --> Research

Green: active and compounding. No fake “shipped to enterprise” labels.

What Worked

Writing as accountability. Publishing forced clarity on what I actually believed vs what sounded good in conversation.

Depth over breadth. RAG failures, ViT deployment tradeoffs, and LLM hype skepticism are related skills. Jumping between them as if they were one company was not — and I was not running a company anyway.

What Failed

Pretending parallel company narratives. Early 2024 drafts sometimes sounded like I had a board and payroll. I did not. I had exams, then a campus ID card.

Sleep. December productivity was fine. August decision quality was not.

Conference FOMO. Every event I skipped, I did not miss anything important.

Numbers I Will Share

Lessons for 2025

  1. Ship companies when incorporated, not when theorized
  2. College schedules are real constraints, not excuses
  3. Kill side project features in November, not March after sunk cost rationalization
  4. Keep human escalation paths in any high-stakes ML deployment you might build later

Looking Forward

2025 was about AI economics shifting again (DeepSeek), side projects getting serious, and deciding which ideas become real commitments. Not a third narrative for the sake of looking busy.

Happy new year. Back to work Monday.

--claps