I own systems end to end, from first design to production.
Logistics · EV fleets · Crypto infrastructure
2 platforms built from scratch · ex-Nethermind · Zcash Zebra security fixes

An EV fleet platform tracking 3,000+ electric vehicles and 3.7M events a day. I was brought in to find and fix its hardest problems.

Almost all of this lives in BatteryFlow's private GitLab, so the contribution graph below barely sees it.

• Route planning in 2 weeks. A feature the team had never cracked, now powered by an in-house routing engine for all of India that handles 2,000+ routes a second on just 2 CPUs, with 1,000x room to grow at a fixed cost.
• Data operators can trust. Revived a partner feed that had dropped 613k packets, zeroed 705k impossible-speed readings and ended 6.8M junk errors a week sent to the partner, so alerts run on clean data.
• Fleet onboarding. Led onboarding for a national battery-swap network and a new OEM partner, then architected self-serve onboarding so business teams can launch new fleet customers without engineering in the loop.
• A second product in 18 days. Single-handedly built BatteryFlow Rental for riders and warehouses (181 commits, 38 PRs) with verified payment webhooks, OTP login and tenant-isolation tests that fail the build.

TypeScript Kafka Redis PostgreSQL BigQuery Apollo GraphQL Kubernetes GCP

An Indian freight marketplace for shippers, carriers and drivers. The founders brought the vision for scale, cost, legal and tax compliance. I designed, built, hosted and ran every layer of it, from zero to its first paid trip in production in 4 months.

• Quotes that cover the real cost of a trip. Replaced flat pricing that underpriced fuel by 38% with a 4-layer pricing engine, built in one day, that matches the fleet's own cost model to 0.5% median error.
• Delivery restored. Deploys had silently failed for 3 weeks. I rebuilt CI/CD with health probes, then shipped 115 production deploys in 31 days across 7 microservices, authoring 125 of the first 126 PRs.
• Accounts protected before launch. Closed 34 review findings in 2 weeks (11 PRs in a day), including open writes to the pricing tables behind every quote and reset tokens that worked as full logins.
• Live truck tracking inside Google Maps' free tier. One cached ETA call per trip every 45 seconds, so the mapping bill stays flat however many people watch a truck at once.

TypeScript Fastify Next.js PostgreSQL Redis GCP CI/CD

Custom-built swap execution for Janction DEX, a Uniswap V3-style exchange on JASMY Chain, around the client's Japan-specific market requirements that standard DEX logic could not handle, plus its Subgraph analytics.

Joined the core-infrastructure team behind one of Ethereum's leading execution clients to work on Juno, Nethermind's Go full node for Starknet, across P2P networking, chain sync and the JSON-RPC API. Chased a P2P bug where nodes never rejoined peers after going offline and briefed the core maintainers.

• Wrote both security fixes in Zebra v6.2.2, the full node of a $26B+ privacy network, with both merged in 31 hours. One closed a shell-injection path so a malicious log line can no longer run commands on a node operator's machine (#11050). The other stopped the Elasticsearch password leaking into the startup config dump (#11051).
• 4 merged PRs in Zebra, and one of the 21 named contributors to Zebra v6.4.0.
• Merged work in librustzcash and ZecHub, including a fix that brought 45 missing hackathon projects back to zechub.wiki.

• MEV Shield · Cut the modeled sandwich-attack loss on a $660K (250 ETH) swap from $35.9K to under $1K, a 97% reduction, by simulating the attacking bot on live Ethereum data and picking the cheapest safe route. Live demo
• Earshot · Watch and steer Claude Code agent sessions across machines from one page, end-to-end encrypted so the relay never sees code or prompts, with guest links that can watch but never issue commands. Published on npm with zero runtime dependencies.
• Streaming ingest pipeline · Crash-safe ingestion with Redis Streams consumer groups, 50 async workers and one atomic rate limiter shared across every replica, so scaling out never breaks the upstream quota.
• GKE on Terraform · Infrastructure as code for Google Kubernetes Engine, with Kustomize overlays, smoke tests that roll back automatically, disruption budgets and anti-affinity.

• Languages and frameworks · TypeScript, Python, Go, Rust, SQL, Solidity, Node.js, Fastify, React, Next.js, Apollo GraphQL
• AI engineering · LLM agents, tool calling, RAG, vector search, MCP servers, Claude Agent SDK, structured outputs, guardrails
• Distributed systems · Event-driven microservices, Kafka, Redis, PostgreSQL, BigQuery, Kubernetes, GCP, Terraform, CI/CD
• Web3 and security · Ethereum, smart contracts, DeFi, MEV, Zcash, Starknet, P2P networking, application security, tenant isolation

IIT (ISM) Dhanbad · Integrated Master of Technology · class of 2026 · Mumbai, open to relocation (US, UK, UAE)

Got a problem where the hard part is figuring out what is actually broken? I would love to hear about it.