agentic ai · stablecoin infrastructure · multi-chain backends

Systems that scale,
settle, and self-heal.

Senior Backend Engineer building production agentic AI systems and stablecoin payment infrastructure. At Avvio, a US neobank, I architected a 36-module NestJS backend spanning EVM, Solana, and Bitcoin with 7 fiat rails across 5 currencies — verified by 335 passing unit tests. Every service ships with server-side security invariants, idempotency keys at every execution boundary, and cron-driven reconciliation that converges state after failures. Before that: multi-agent LLM backends on Google ADK, MCP, and Vertex AI at a Singapore AI lab, and enterprise-scale RAG at Jio.

36 production NestJS modules shipped
3+7 chains + fiat rails · 5 currencies
335 unit tests · green build
p95 latency improvement at scale

Production patterns, runnable in the browser.

Two systems I run in production, simulated here: a distributed transaction saga with compensating rollback, and a self-healing reconciliation daemon. Run them and watch the state machines converge — this is the engineering behind every bullet point on this page.

guest@ritesh-system: ~ — production logic
> curl -s https://api.ritesh.dev/status | jq .
{ "name": "Ritesh Khamitkar", "title": "Senior Backend Engineer · Agentic AI & Stablecoin Infrastructure", "experience": "4+ years · production-grade", "agentic_ai": { "orchestration": ["MCP server authoring", "Google ADK", "LangGraph", "multi-agent state machines", "tool calling"], "production": ["eval harnesses", "LLM-as-a-judge", "tracing (OTel)", "guardrails", "cost-per-task budgeting", "model routing"], "retrieval": ["RAG pipelines", "LlamaIndex", "Pinecone", "Neo4j", "Mem0"] }, "stablecoin_infra": { "chains": ["EVM / Base L2", "Solana", "Bitcoin (UTXO/PSBT/Taproot)"], "rails": ["ACH", "Wire", "SEPA", "CLABE", "PIX", "Lightning", "Card"], "wallets": ["Turnkey MPC", "Policy Engine", "Paymasters", "EIP-7702"], "settlement": ["Relay intents", "LI.FI", "self-healing reconciliation"] }, "backend": ["NestJS", "FastAPI", "Kafka", "PostgreSQL", "Prisma", "Redis", "gRPC"], "infra": ["AWS (EC2/RDS/S3)", "GCP Cloud Run", "Docker", "k8s", "GH Actions"], "proof": "36 modules · 335 green tests · 4× p95 latency win", "status": "available_for_technical_evaluation" }
> node saga_engine.js --execute-swap

// Distributed multi-chain swap saga (NestJS logic, real recovery paths)

IDEMP_CHECK
INITIATED
ESCROWED
SOL_RPC_FAIL
ROLLED_BACK
// State machine idle. Click "run" to test the distributed transaction saga.
> python reconcile_daemon.py --detect-drift

// Self-healing ledger reconciliation & finality daemon (PostgreSQL + Base L2)

SCAN_POSTGRES
STUCK_DETECTED
QUERY_CHAIN_RPC
CONFIRMATIONS (0/12)
RECONCILED_SETTLED
// Daemon sleeping. Click "trigger" to simulate post-restart database recovery.

Six domains, shipped end to end.

Distributed-systems fundamentals first — agentic AI and stablecoin infrastructure on top. Every item below has run in production.

Agentic AI & LLM Production
  • Multi-agent orchestration · Google ADK · LangGraph · LangChain · OpenAI Agents SDK
  • MCP (Model Context Protocol) server authoring · tool calling · agent state machines
  • Eval harnesses · golden datasets · LLM-as-a-judge · regression gating in CI
  • Agent observability: tracing (OpenTelemetry, LangSmith, Langfuse) · cost-per-task budgeting · model routing
  • Guardrails · prompt versioning · context window management
  • RAG pipelines: chunking → embedding → vector index (Pinecone, pgvector) → reranking · Mem0 · LlamaIndex · Neo4j knowledge graphs · Text-to-SQL
Stablecoin & Payments Infrastructure
  • Stablecoin rails (USDC, USDT, EURC) · multi-chain routing · cheapest-fee selection
  • Fiat on/off-ramp: Bridge.xyz (ACH, Wire, SEPA, CLABE, PIX) · virtual accounts · payouts
  • On-ramp aggregation: Coinbase CDP, MoonPay, Swapped · fail-open quoting
  • KYC orchestration: Sumsub, Persona, multi-partner submission pipelines
  • Ledger & reconciliation · durable work queues · deterministic idempotency keys
  • Webhook security: HMAC signatures · timestamp replay protection · idempotent processing
Blockchain & Smart Contracts
  • Solidity · Hardhat · Foundry · gas optimization · CREATE2 deterministic deployment
  • ERC-20 / ERC-721 / ERC-4626 vaults · OpenZeppelin (ReentrancyGuard, Pausable, SafeERC20)
  • Base L2 · keccak256 commit-reveal escrows · on-chain payment links
  • Turnkey MPC custody · policy engine · EVM & Solana paymasters · gasless transactions
  • Account abstraction: EIP-4337 · EIP-7702 delegation detection
  • Bitcoin: UTXO selection · PSBT construction (Taproot) · fail-closed output guards · mempool broadcast
Distributed Systems Fundamentals
  • CAP & PACELC trade-offs · consistency models (strong / eventual / causal)
  • Saga pattern · compensating transactions · outbox pattern · CQRS · event sourcing
  • Idempotency keys · exactly-once semantics · deduplication at every boundary
  • Cross-chain intents (Relay, LI.FI) · quote lifecycle · state convergence
  • Latency vs throughput reasoning · back-of-envelope estimation
  • Leader-leased work queues · single-writer safety · guarded state transitions
Scaling, Reliability & Observability
  • Circuit breakers · bulkheads · backpressure · rate limiting · retry with jitter
  • Dead-letter queues · poison-message isolation · CRON reconciliation daemons
  • Structured logs · distributed tracing (OpenTelemetry) · SLI/SLO design · error budgets
  • Health checks · readiness/liveness probes · graceful shutdown · zero-downtime deploys
  • Connection pool sizing · multi-level caching (in-memory → Redis → DB) · TTL & invalidation
  • Failure-mode analysis · graceful degradation · fail-open vs fail-closed decisions
Backend, Data & DevOps
  • NestJS · FastAPI · TypeScript · Python · Node.js
  • Kafka consumer groups · Redis · gRPC · WebSockets · REST
  • PostgreSQL · Prisma · BigInt-precision financial math · schema migrations
  • Docker · Kubernetes · GitHub Actions CI/CD
  • AWS (EC2, RDS, S3, Secrets Manager, CloudTrail) · GCP (Cloud Run, Vertex AI)
  • OAuth2 · JWT auth guards · HashiCorp Vault · RBAC · secret rotation

The fundamentals, enforced.

Patterns I audit my own code against before anything ships. Each one has caught a real bug in production — this is the difference between code that runs and code that survives.

01

Algorithmic complexity

Hunt O(n²) and redundant passes — filter().includes(), nested loops, sort-then-slice. Reach for Map/Set lookups, partial sorts, and single-pass reductions. Profiling the loop before optimizing it.

02

Concurrency & backpressure

Kill sequential awaits that should be Promise.all / asyncio.gather. Bound parallelism with semaphores and pools. Replace sync I/O in async contexts. Backpressure every queue before it back-floods the producer.

03

Caching & invalidation

Memoize expensive computations, TTL every cache entry, evict LRU. Multi-level read path: in-memory → Redis → DB. SWR on the client, Cache-Control + ETag on the wire. Never cache time- or session-scoped data.

04

Error resilience

Timeouts on every external call (AbortController / client timeouts). Exponential backoff + jitter on transient failures. Circuit breakers around every dependency. Empty catch is a bug; graceful degradation beats hard failure.

05

Database & queries

Eager-load to kill N+1, SELECT only the columns you need, index the columns you filter and join on. Bulk operations (bulkCreate, executemany) over per-row writes. Connection pools sized to concurrency, never per-request.

06

Observability

Structured JSON logs with a correlation ID propagated end-to-end. No console.log in hot paths. Pre-aggregated metrics in loops. Distributed traces on every request. The question is never "is it slow?" — it's "where, for whom, since when?"

Three roles. One discipline.

Stablecoin payments infrastructure from scratch at a US neobank, agentic AI at a Singapore lab, and enterprise-scale ML at Jio. Same engineering standards at every scale.

case 01 · founding engineer (blockchain & backend) Avvio
Nov 2025 — Present · Remote (US)
  • Architected a 36-module NestJS stablecoin payments backend from scratch — modular Service–Handler pattern with bounded contexts per chain domain (EVM, Solana, Bitcoin), stateless services with zero shared mutable state, verified by 335 passing unit tests.
  • Engineered gasless transaction infrastructure across EVM and Solana (Turnkey MPC custody, Gas Station, paymasters, Alchemy RPC, ethers.js); enforced server-side wallet resolution on every deposit flow and a fee-payer-only paymaster validator that structurally blocks all drain vectors.
  • Developed, tested, and deployed the AvvioPaymentLinkV2 Solidity contract on Base L2 — dual escrow flows (request & send links), keccak256-hashed claim secrets, CREATE2 deterministic deployment via Hardhat, with OpenZeppelin ReentrancyGuard, Pausable, and SafeERC20.
  • Built native Bitcoin infrastructure — UTXO selection, PSBT construction with Taproot support, fail-closed output guards, direct mempool broadcast — plus an @username transfer router that auto-selects the lowest-fee chain (EVM vs. Solana) for multi-chain USDC.
  • Integrated multi-partner fiat and KYC orchestration: Bridge.xyz on/off-ramp (ACH, Wire, SEPA, CLABE, PIX across USD/EUR/MXN/BRL/GBP), Sumsub/Persona identity verification, and Paytrie/Seismic regional rails — all webhooks HMAC-verified with replay protection and idempotent processing; deployed Turnkey embedded business wallets with multi-user RBAC and on-chain policy engine (spending limits, vendor allowlists).
  • Shipped self-healing reconciliation: cron-driven reconciler services on durable, lease-based work queues with deterministic idempotency keys — plus ERC-4626 yield vault integrations (YO Protocol, Blend.money) with cross-chain stablecoin consolidation over Relay intents. Deployed on AWS (EC2, RDS PostgreSQL, S3, Secrets Manager, CloudTrail) with GitHub Actions CI/CD, multi-stage Docker builds, and PM2.
case 02 · founding engineer (senior AI & web3 backend) Qurios AI
Jan 2025 — Oct 2025 · Remote (Singapore)
  • Architected production agentic AI backend: multi-agent orchestration with Google ADK, MCP server, and Mem0 vector memory; deployed Vertex AI Reasoning Engine agents on GCP Cloud Run with horizontal autoscaling and graceful shutdown on SIGTERM — no cold-start latency spikes.
  • Designed multi-stage RAG retrieval pipelines (LlamaIndex, Pydantic, async ThreadPoolExecutor) with parallel LLM calls; cut API p95 latency 4× via asyncio parallelization, Firestore read caching with TTL invalidation, and connection pool sizing tuned to concurrency profile.
  • Built high-throughput NestJS microservices with Kafka consumer groups ingesting multimodal streams (Twitter, Telegram, YouTube, Substack, PDFs, audio) for automated AI personality creation; dead-letter queues for poison messages; CRON-based ETL and gRPC (BloomRPC) inter-service communication.
  • Integrated X402 protocol with Coinbase CDP Facilitator for on-chain micropayments; deployed Base MiniApp end-to-end with LlamaPay subscription streaming, webhook listeners, and Privy wallet auth; built Neo4j knowledge graphs with Pinecone vector embeddings for contextual retrieval and Text-to-SQL refinement models.
case 03 · devops & AI/ML engineer Jio Platforms Limited
May 2023 — Dec 2024 · Navi Mumbai, India
  • Architected production AI backend for Jio Brain — multi-agent orchestration with FastAPI, RAG pipelines (Pinecone vector memory + Neo4j knowledge graphs), and distributed anomaly detection at enterprise telecom scale; designed for horizontal scaling with stateless Cloud Run services and Kafka-backed event ingestion.
  • Built high-throughput Kafka ingestion microservices with consumer group offset management, CRON-based ETL pipelines, async Cloud Run batch jobs, and gRPC inter-service communication; implemented dead-letter queues for poison-message isolation and CRON reconciliation for idempotent re-processing.
  • Developed production RESTful APIs with OAuth2 authentication and HashiCorp Vault secret management; automated MLOps using GitHub Actions and Kubernetes (rolling deployments, liveness/readiness probes, zero-downtime updates) for continuous model deployment pipelines.
  • Implemented distributed model training on GCP with TensorFlow Extended (TFX) and Kubeflow; applied Bayesian Optimization and Grid Search hyperparameter tuning; reduced API p95 latency 4× through async parallelization, Firestore caching, and targeted connection pool configuration.

Four problems where design carried the risk.

The interview question every senior role asks — answered with production systems. Each tile follows the same arc: the problem, the constraint that made it hard, the design decision, and the outcome.

problem 01 · security boundary

Untrusted deposit destinations

Problem: any deposit flow that accepts a destination address from the client is one compromised app away from redirecting user funds. Constraint: three chains, multiple on-ramp providers, one invariant. Decision: destination addresses are always resolved server-side from the internal Wallet table (Turnkey MPC sub-org accounts) — the client never supplies one. Outcome: the entire class of redirect-deposit attacks is structurally impossible, not merely validated against.

  • Wallet addresses read server-side from User → Wallet
  • Client never sends a destination for any deposit flow
  • EVM / Solana / Bitcoin addresses derived from Turnkey sub-org accounts
problem 02 · on-chain safety

A gas sponsor that cannot be drained

Problem: a Solana paymaster signs user transactions — and any instruction referencing its key can debit it. Constraint: instruction blocklists always lag new attack vectors. Decision: a strict structural validator — the paymaster must sit at account index 0, address lookup tables are rejected (they hide accounts), and no instruction may reference the paymaster key at all. Outcome: every drain vector (SystemProgram.transfer, createAccount, SPL, CPI) is blocked by a three-rule invariant instead of an ever-growing blocklist.

  • Fee payer locked to account index 0
  • ALTs rejected — no hidden account resolution
  • Blocklist-free drain prevention via index invariant
problem 03 · resilience

Multi-provider quoting that never fails whole

Problem: on-ramp providers (Coinbase, MoonPay, Swapped) each fail differently — errors, unsupported countries, below-minimum amounts. Constraint: one flaky provider must never take down the deposit screen. Decision: parallel quote fetching with fail-open semantics — a failing provider is dropped, not fatal — cheapest-fee sort, USDC→USDT stablecoin fallback, and session tokens minted at commit time so they cannot expire mid-comparison. Outcome: users always see the best available quote, with a "minimum amount" floor instead of a dead end.

  • Parallel quoting across N providers, nulls dropped silently
  • Fail-open: errors exclude the provider, not the request
  • Commit-time token minting — sessions never expire mid-quote
problem 04 · distributed state

State that converges after every failure

Problem: clients stop polling, pods restart, and cross-chain transfers get stranded mid-flight. Constraint: in-memory retry loops die with the process. Decision: durable, lease-based reconciliation queues driven by cron — deterministic idempotency keys, guarded state transitions (PENDING/EXECUTING-only writes), and chain-RPC finality checks. Fragmented balances consolidate across chains before high-value sends. Outcome: abandoned transfers converge to a terminal state with zero manual intervention — the exact pattern simulated in the terminal above.

  • Durable work queues with leases — reconciliation survives restarts
  • Idempotent, guarded writes stay consistent with webhooks and live polling
  • Cross-chain balance consolidation before high-value sends
36 NestJS modules shipped
3 chains · evm · solana · bitcoin
7 fiat rails · 5 currencies
335 unit tests · green build

Coding agents as a force multiplier — with discipline.

I use Claude Code, Cursor, and Kimi Code as primary development interfaces — the workflow modern startups now hire for explicitly. The multiplier comes from the harness around the agent, not the agent alone: specs before prompts, review gates on every merge, and tests that verify the output.

01

Spec-first prompting

Architecture, interfaces, and invariants are designed before any agent writes code. The agent implements against a written spec — module boundaries, DTO contracts, error semantics — so generated code lands inside a deliberate design, never instead of one.

02

Harness design

Repo rules, custom MCP tools, and per-module documentation give agents the context to produce code that matches house conventions. The 36-module Avvio backend maintains agent-readable docs per module — the same docs that onboard human engineers.

03

Verification gates

Nothing merges on the agent's word. Every change passes human review, the full unit-test suite (335 green at Avvio), and lint/build gates in CI. Security-sensitive paths — paymaster validation, PSBT output guards, webhook verification — get line-by-line manual review.

04

Tool selection by task

Cursor for IDE-native refactors and navigation-heavy edits; Claude Code for multi-file backend migrations and long-horizon tasks; Kimi Code for high-volume implementation against tight specs. Knowing when not to use an agent is part of the skill.

05

Measured output

The result is throughput without debt: a 36-module payments backend across 3 chains and 7 fiat rails, shipped by a small founding team on startup timelines — with the test coverage and security invariants of a much larger organization.

06

Agent systems, both sides

I build agentic systems (MCP servers, multi-agent orchestration, eval harnesses) and I build with agentic tools daily. Each side sharpens the other — production agent reliability patterns come directly from operating these tools under real constraints.

Academic foundation.

Education

M.S. Artificial Intelligence BITS Pilani Apr 2023 — Mar 2025 · India
B.E. Information Technology Pune University Jun 2019 — May 2023 · India

Open channels.

Available for technical evaluation, founding roles, and architecture reviews. Email is fastest; LinkedIn for the long-form context.

pre-engagement technical evaluation

Evaluate before you commit.

Open to architecting and shipping one production-grade service, smart contract, or AI agent integration as a no-obligation technical evaluation (1–2 weeks) before any formal commitment. Pick a real feature from your backlog — I will deliver it with production-grade system design, tests, and documentation, so you can judge execution speed and code quality directly.

book a call →