$whoami

raviraj kumar

building in stealth · founder · builder

two bets. neatskills - outcome-driven skill learning, live mentors + ai tutor, fraction of bootcamp cost. quantum finance - hybrid qaoa + amplitude-est mc libs for asset managers, ships now, scales w/ hardware.

$cat right-now.md

neatskills: cohort 1 closed (39 pilots, 20%+ referral, 3 universities). cohort 2 opening. quantum finance: qaoa allocator + ae-mc q3 2026, open lib + paper q4 2026. design-partner slots open.

last updated · 2026-06-02

neatskills.in - outcome-driven skill learning. live mentors, adaptive resources, ai tutor between sessions. deep dive →

• - problem: bootcamps cost $5-15k, churn 60%+, sub-40% placement. moocs are worse.
• - wedge: live mentors + adaptive resources + ai tutor between sessions. cohort intimacy w/ ai leverage, fraction of bootcamp price.
• - ai tutor: tool-calling agent scoped to learner's current module + portfolio. drills, code review, hints between mentor sessions.
• - adaptive assessments are a tiny eval engine. weekly skill-delta on held-out tasks the learner hasn't seen. mentor feedback closes the loop on whatever the eval misses.
• - traction: 39 pilots, 20%+ unprompted referral, 3 universities in pipeline. first to complete founders ashram cohort 1.
• - why now: ai lets one mentor pool serve 10x learners without diluting outcomes. window open.

hybrid libs for risk + pricing problems classical compute can't reach economically. ships into existing quant pipelines now. deep dive →

• - problem: $50t+ aum industry. classical mc for risk + pricing eats compute, undercounts tail.
• - wedge: hybrid libs - qaoa allocator, amplitude-estimated mc, variational kernels. quantum where it earns its keep.
• - numerics: amplitude estimation targets near-quadratic speedup on monte carlo; var/cvar in seconds where classical pricing pipelines need minutes.
• - hardware ready now: ibm heron (133 qubits), ionq forte, quantinuum h2 (56 qubits, all-to-all). enough for production poc on tail-risk + portfolio cuts.
• - deliverable: open sdk + paid runtime. paper + lib target q4 2026. design partners onboard before code is final.
• - adjacent track: llm finance agents evaluated against classical ground truth. private hold-out sets, agent-level scoring, no train-on-test.

• - neatskills · 39 pilots · 20%+ unprompted referral · 3 universities in pipeline
• - neurallearn · rank 10 sf regionals · $1k prize · eazo.ai global hackathon
• - founders ashram cohort 1 · first to complete (program record)
• - delta-founder-directory · 104 startups · 2.4k visits in 4d · shipped in a weekend
• - iit delhi · top 1000 india (2025)
• - iit (ism) dhanbad · top 10 finalists, aavishkar 2.0 (2022)

• - neatskills pre-seed intros (edtech + ai infra)
• - quantum finance design partners (quant funds, insurers, family offices)

• [01]
  neurallearnshipped

  built at eazo.ai global hackathon. rank 10 sf regionals, $1k prize. shipped in 48h.
  $1k · rank 10
• [02]
  provoke developerslive

  legal entity housing every venture + idea i ship — neatskills, quantum finance, and more. founded mar 2022, 4y+ running.
  4y+
• [03]
  stealth iot wearablelive

  head of product. accessibility hardware, under nda.
  feb 2026
• [04]
  delta-founder-directoryshipped

  delta ii directory. 104 startups, 2.4k visits in 4d. shipped over a weekend.

• - eazo.ai global hackathon · rank 10 sf regionals · $1k prize (neurallearn)
• - first to complete founders ashram cohort 1 (program record)
• - iit delhi · top 1000 india (2025)
• - iit (ism) dhanbad · top 10 finalists, aavishkar 2.0 (2022)
• - the residency · delta ii (oct-nov 2025)
• - startup mahakumbh 2025 · silver delegate

• - usable in a week > perfect in a year
• - unprompted referrals > any market research
• - evaluation rigor is the moat - in edtech and in ai
• - finance llm eval is the highest-leverage benchmark problem right now. real frictions, no leakage, agent-level.

• ? how do you build a finance agent benchmark that survives a year of model releases without leaking?
• ? what's the right held-out task when the curriculum itself adapts?
• ? when do quantum primitives stop being scientific demos and start being procurement line items?
• ? what does fair evaluation look like when the evaluator and the model share training pre-history?

• currently · thinking, fast and slow · daniel kahneman
• next · designing data-intensive applications · martin kleppmann

• ›
  the almanack of naval ravikant

  specific knowledge + accountability + leverage. long games, long-game people.

  · eric jorgenson
• ›
  the cold start problem

  atomic networks first. you don't skip to escape velocity, you earn it.

  · andrew chen
• ›
  you can make it happen

  execution under constraint. systems > heroics.

  · anil swarup

$contact --all