IIT Bombay student building applied machine-learning, quantitative-finance, analytics, and operations projects. This profile emphasizes reproducible evidence: each highlighted claim links to code, data provenance, tests, or a report. Simulated and real datasets are labeled explicitly.
Live application | Public strategy case
- Co-founded an MSME financial-intelligence product deployed with AWS EC2 and RDS inside a VPC, with Secrets Manager used for application credentials.
- Built transaction intelligence over 81,813 mixed anonymized and generated records across 28 categories.
- Ran one-month free pilots with 10+ MSMEs; 10+ returned after initial use.
- No paid-customer or revenue claim is made for FlowFinance.
| Project | Evidence |
|---|---|
| Mistral-7B QLoRA | 1,200-step run, rank/target-module ablations, held-out perplexity reports |
| CIFAR-10 DDPM | 71.03M-parameter U-Net, 50K-sample evaluation, FID 10.0958, IS 8.7801 |
| Mini-GPT | Decoder-only model, KV-cache benchmarks, ablations, LoRA adapters, model card |
| ML Serving and Monitoring | FastAPI, Docker, latency reports, PSI drift checks, tests |
| Project | Data status and evidence |
|---|---|
| BTC Adaptive Grid Research | Live Binance WebSocket paper engine, persistent state, bid/ask fills, and 40,536-candle holdout evidence |
| OpenBB Macro Portfolio Research | 16,919 real records across seven ETFs; monthly walk-forward optimization, costs, bootstrap intervals, and dashboard |
| Factor Modeling | Adjusted prices for 49 real US equities; PCA and classical-factor diagnostics |
| Regime-Aware State-Space Research | Real SPY/QQQ/TLT prices; Kalman state estimation, Gaussian HMM regimes, turnover costs, and honest passive-benchmark limits |
| Cointegration Pairs | Nine real US ETFs; Johansen, Kalman, HMM, cost-aware backtests |
| Market Making | Simulated LOB; HJB-derived quotes, Monte Carlo stress tests, PnL attribution |
| HFT Signal Research | Synthetic 100K-event LOB; executable-edge and transaction-cost diagnostics |
| Options Pricing | Heston, Black-Scholes, Monte Carlo/QE, Greeks, calibration tests |
| Project | Evidence |
|---|---|
| Manufacturing Quality Analytics | Simulated 52K-batch operations study plus real UCI steel-fault benchmark and dashboard |
| Credit Risk Lifecycle Analytics | Synthetic portfolio plus real OpenML benchmark, SQL approval simulator, drift/fairness checks |
| Supply-Chain Network Design | Facility-location MILP, scenario stress tests, RAG decision support, Streamlit dashboard |
| SQL Operations Analytics | DuckDB KPI layer, cohorts, query profiling, SLA governance |
| Demand Forecasting | Rossmann forecasting, model comparison, safety-stock and service-level decisions |
| Experimentation | Funnel analysis, A/B testing, power/MDE, CUPED and heterogeneous effects |
Funded Nation and Evore Labs work involved internal prototypes. Public claims are limited to aggregate, employer-verifiable outcomes. No proprietary data, customer records, strategy logic, private screenshots, or confidential source code is published. Sanitized case studies will be added only after written employer approval.
- Real, anonymized, generated, and simulated data are identified separately.
- Negative or rejected research results remain visible when they are the honest result of leakage-safe and cost-aware evaluation.
- Repository metrics should be reproducible from committed reports or commands.
- Paper trading and exchange Testnet activity are not described as live-money trading.