Final-year dual-degree student at BITS Pilani (B.E. Mechanical + M.Sc. Chemistry, 2027), interning in Client Success & Program Delivery at Goodera, Bengaluru.
I build agents and growth systems that run in production: with guardrails, evals, and impact measured through experiments.
📫 f20220970@pilani.bits-pilani.ac.in · LinkedIn
| What | Live link | What it shows |
|---|---|---|
| LiveOps agent | Daily eval page | Tunes a simulated puzzle game's difficulty: A/B tests each change, ships or rolls back inside retention and revenue guardrails, logs every decision. Re-evaluated every day on GitHub Actions; the Gemini ops note is number-checked. |
| AI inbox triage agent | Daily eval page | Claims guardrail → Gemini → local fallback → confidence router. On unseen wording: 20% → 72% accuracy vs the local model. Refuses to publish when the LLM fails. |
| Growth Console | Streamlit app | Founder planning tool: move growth levers and watch ARR, cash runway and Monte Carlo risk update. |
| Call Insights | Live demo | FastAPI + Next.js app turning English/Hinglish calls into insights, with an ops agent. 94% intent accuracy vs 62% baseline. |
| Project | Highlight |
|---|---|
| game-ua-budget-agent | Spends an ad budget on predicted D30 player value with Thompson sampling and guardrails: +10.6% revenue vs even split; caught a bidding incident in 1 day |
| retention-uplift-engine | Uplift targeting 113 vs 69 extra orders per 1,000 vs churn-risk targeting; Kaplan–Meier; propensity-matched loyalty ROI |
| marketplace-pricing-experiments | Switchback carryover bias (37%), elasticity with clustered SEs, CUPED (−87% variance), constrained price optimiser |
| dynamic-pricing-bandits | Shared-model Thompson sampling: 98.4% of oracle, 60% less regret than an A/B test |
| dispatch-matching-sim | Hungarian batching + forecast-led repositioning: fulfilment 64% → 84% with half the fleet |
| Project | Highlight |
|---|---|
| soma-brokerage-ops-lab | Insurance brokerage ops: speed-to-lead, carrier routing, renewals, AI triage eval · founders' memo |
| arr-growth-model | 24-month ARR and cash model in live Excel, matched to Python; Monte Carlo |
| revenue-pipeline-forecaster | Deal-level Monte Carlo within 3.8% vs 83% error for stage-weighted CRM forecasts |
| porter-category-growth-lab | Pricing × fulfilment, partner acquisition and M1 retention, driver-support redesign |
| qcommerce-retention-sql | 9 SQL analyses on 3.35M Instacart orders, growth experiments, FirstClub teardown |
| noon-funnel-growth · eurekaforbes-service-insights · nobroker-strategy-case | Funnel, service-analytics and strategy case studies |
| finance-recon-controls · credit-risk-model-monitoring · trip-data-pipeline | Reconciliation controls, credit-risk monitoring, ETL over 5.97M trips |
Company-specific projects use simulated or public data and say so in their READMEs; they show method, not company facts.
Tools: Python · SQL · NumPy/pandas/scikit-learn · Gemini API · FastAPI · Streamlit · GitHub Actions · Excel · Claude Code