SUKHMAN VIRK

• Sole engineer on a space-object custody system for a space-domain-awareness company, shipped in under 4 weeks: Django and React over ~27,000 catalogued objects, each with an accountable owner.
• Built its RAG assistant on AWS Bedrock Knowledge Base with Cohere embeddings, semantic chunking, and metadata-filtered OpenSearch Serverless vector search.
• Designed an MCP server via FastMCP exposing five role-gated tools to Bedrock agents, enabling hybrid retrieval across PostgreSQL and vector stores with permissions enforced at the tool layer.
• Led a team of 4 engineers building Minfy's reusable satellite earth-observation platform on AWS.
• Built the natural-language geospatial search for a multi-tenant weather and satellite data marketplace: an LLM parses the request and PostGIS does the spatial retrieval, scoped to each tenant's plan.
• Built the multi-tenant RBAC admin console and Go APIs for an IoT access-control platform for a lock manufacturer, isolating data per tenant on shared infrastructure.
• Shipped a secure file-management service in Flask with KMS encryption, sustaining 10,000+ daily requests at 99.99% uptime.
• Led greenfield builds and on-premises to AWS migrations with automated failover, circuit breakers, and zero-downtime cutovers, instrumenting CloudWatch metrics to trace failures across service boundaries.

• Built a C++ WebRTC peer-to-peer application for a Fortune 500 client moving images, video, and messages at up to 4K resolution.
• Automated ICE negotiation via Google STUN, cutting connection setup time by 90% and end-to-end media latency by 30%.

• Built a full-stack AES-256 cryptographic applet in Next.js and Java protecting 50,000+ user records with zero post-launch vulnerabilities.
• Automated decryption-key retrieval via AWS KMS and integrated Okta 2FA, eliminating manual key rotation across the platform.

A two-sided market maker live on Kalshi BTC hourly binaries, quoting across the strike ladder and requoting on every order-book event in under 50ms, across 100+ markets with $10K+ in traded volume.

• Suppresses quoting during information events via a toxicity-triggered participation gate under asymmetric hysteresis, the largest single driver of realized PnL in the system, ahead of pricing precision.
• Estimates fill probability with a censoring-aware survival model trained on 1M+ order events over queue, depth, and book-state features.
• Prices adverse selection per fill with a gradient-boosted markout model under Bayesian regularization, subtracting expected markout from captured edge rather than booking the full spread as profit.
• Sets spread, size, and hold horizon from an explicit profit optimizer maximizing expected capture net of adverse selection, with inventory carried as a quadratic variance penalty.
• Centers quotes on a reservation price in log-odds space rather than the Avellaneda-Stoikov closed forms, whose spread and skew terms degrade on binary contracts near the tails.
• Production stack in Python: signed exchange auth, Redis hot cache with durable fill history, amend-first order management, and reconciliation crediting only verified fills to prevent phantom inventory.

A full-stack quant research platform with a daily ETL pipeline ingesting equities, options chains, and Treasury yields.

• Black-Scholes IV inversion (Newton-Raphson) and an options Greeks landscape, with cubic interpolation for smooth implied volatility surfaces.
• LRU caching on iterative numerical models cut per-query runtime by 4x.
• Published cuda-kernels to PyPI: CUDA-accelerated autocorrelation and sum reduction running 91% faster than NumPy.

App repo · PyPI · Kernels repo

A Dixon-Coles team-strength model, a multi-period MILP squad optimiser, and a backtest harness that gates every change on held-out seasons. The MILP cleared its pre-registered bar on a held-out season (+56.3 points over greedy, 5 of 6 seeds ahead). Benchmarks weekly against managers with repeated top-1% finishes, read live from the FPL API.

End-to-end NLP pipeline on 14,600 tweets: TF-IDF features with SVM and Logistic Regression for 3-class sentiment, K-means with PCA clustering validated by Silhouette scoring, and LDA topic modeling to surface latent complaint themes.

Repo

Real-time Android app in Java rendering friends' live locations from persistent location streams over a SQL-backed friend graph, with a full Robolectric, JUnit, and Espresso test suite on GitHub Actions CI/CD.

Repo

• Built a QR code check-in system that cut queue wait times by 50% and eliminated unauthorized entry across 10+ events.
• Maintained a website serving 7,000+ monthly visitors.