// projects

Work

Systems I've designed and built, from distributed API gateways to multi-agent AI pipelines.

// experience

Where I've Worked

HSV Digital

Software Engineer

Dec 2025 - Present

Remote

Shipped 8 backend modules for a multi-tenant AI sales-readiness platform, including Sense AI, a RAG-grounded assistant with tenant-scoped auth and citation enforcement that cut unsupported responses from 18% to under 0.3%. Built a longitudinal memory system for sub-50ms competency retrieval, 40+ backend-for-frontend APIs, and an event-driven notification service running at 99.9% delivery reliability. Reduced API cold-start time 93%, from 45s to 3s at P99.

TypeScriptNode.jsGraphQLNestJSPrismaPostgreSQLRedisAWSDockerPostHog

Wayground (formerly Quizizz)

Software Engineer

Jan 2025 - Oct 2025

Bengaluru, Karnataka

Built an analytics ingestion pipeline moving 50M+ daily events into BigQuery over Kafka and Pub/Sub, holding end-to-end latency under 30s. Cut search latency 78%, to under 280ms across 10M+ records, by partitioning OpenSearch indexes. Cut quiz response latency 42% (2.1s to 1.2s) across 1M+ daily quizzes with Redis pipelining and GCP autoscaling tuning.

GoPythonKafkaPub/SubBigQueryOpenSearchGCPRedis

Mood India

Software Development Intern

May 2024 - Jul 2024

Remote

Built 25+ REST APIs in Java, C#, and Python/Django, processing 1M+ records monthly and cutting report generation time 65%. Built a C#/.NET pipeline with SQL Server automating 100K+ records monthly, cutting manual prep 70%. Shipped Django/MongoDB APIs for React and React Native clients, and Java services on PostgreSQL/Redis that cut response times 55% for 10K+ monthly users.

JavaC#PythonDjangoMongoDBReact NativePostgreSQLRedis

Serialized-Audio Streaming Platform

A platform for original serialized audio: listeners follow shows season by season and resume to the second across devices, creators publish episodes or generate whole series through an asynchronous AI pipeline that writes the script, synthesises the speech, and packages HLS audio. Eight independently deployable services, one database each, domain events delivered through a transactional outbox.

GoPythonKafkaPostgreSQLRedisFFmpegNext.jsTypeScriptDocker
8 services, database-per-service, outbox-relayed events

Vector Search Engine, Built From Scratch

A vector database written from scratch in Go: HNSW indexing, a crash-safe write-ahead log, mmap-backed snapshots, and tenant-isolated multi-tenancy with scoped API keys. Product quantization and a Raft-sharded cluster exist as tested packages. Benchmarked head-to-head against a real Qdrant instance, not wrapped around one.

GogRPCOpenTelemetryPrometheusDockerPython
13.9K QPS vs Qdrant's 7.6K, 46 MB vs 116 MB
Powers this site's search (press /)

Multi-Tenant API Gateway

Multi-tenant API gateway in Go that routes requests to tenant-registered origins by subdomain. Clerk JWT and scoped API-key auth, Redis-shared rate limiting and response caching, health-aware load balancing, and config pushed from a separate control plane over gRPC, instrumented with Prometheus and OpenTelemetry.

GogRPCPostgreSQLRedisOpenTelemetryDockerNext.jsTypeScript
~70K req/s, +0.5ms p50 over a direct call

OpenAI-Compatible Inference Gateway

Single Go binary that fronts multiple LLM backends, including Ollama, vLLM, or hosted providers like Groq, behind one OpenAI-compatible API. Round-robins across healthy backends, fails over on error, streams responses unbuffered, and semantically caches repeat prompts via NuclaDB so identical requests never hit a backend twice.

GoRedisNuclaDBPrometheusDocker
~880x faster on cache hits

Multi-Agent AI Debugging

Paste your broken API logs, get back what went wrong and how to fix it. Multiple AI agents argue about the diagnosis so you don't have to.

LangGraphGroqFastAPIPythonNext.jsTypeScript
6s avg inference, 30+ reasoning steps

AI-Powered Claude Skill Orchestrator

CLI tool that lets you create, share, and execute reusable AI coding-agent skills without copy-pasting markdown files like it's 2005. Works with Claude, Groq, OpenAI, and Ollama, with sub-100ms local command resolution.

GoRedisPostgreSQLNext.jsTypeScriptDocker
sub-100ms command resolution, 20+ workflow templates

AI Codebase Analysis Agent

Stateless FastAPI service that analyzes GitHub repos with Groq-hosted Llama 3.3 70B: security, architecture, code quality, and performance checks through one API, plus webhook-triggered analysis on every push.

PythonFastAPINext.jsTypeScriptDocker

Self-Serve SaaS Commerce & Contracting

Self-serve commerce app for buying and managing a SaaS subscription, where every purchase produces a real agreement: a generated contract, a signature step, and an audit trail of who did what and when. Handles proration correctly on mid-cycle upgrades and downgrades.

C#ASP.NET CoreGraphQLMongoDBReact

// open source

Contributions

Merged and in-review PRs to production open source projects

contribution activity@Rakshit-gen →

DeepSpeed

Microsoft Research

6 merged · 3 in reviewView repo →
merged
Fixed deadlocks in DecoupledCheckpointEngine#7742

Eliminated indefinite hangs with timeouts & health checks

merged
merged
merged
merged
merged
Fixed autotuner crash when an experiment produces no metric#8421

An OOM'd experiment no longer breaks best-config selection

in review
Kept each param group's beta2 in OneCycle#8751

Adam's default beta2 of 0.999 was silently replaced with 0.99

in review
in review
Rejected an --exclude that leaves no slots to launch on#8755

Clear error instead of an IndexError deep in the launcher

Expo

Expo

2 merged · 1 in reviewView repo →
merged
Fixed dev-server CORS check accepting non-loopback origins#49302

An unescaped '.' in the localhost regex matched any character

merged
Fixed Content-Length in the JS inspector for non-ASCII bodies#49305

Header used string length instead of byte length

in review
Cached evaluated API route modules in the dev server#49259

Module-scope state now survives between dev requests

Accelerate

Hugging Face

merged
Fixed load_accelerator_state restoring only one RNG backend#4217

Resuming a checkpoint restores every saved RNG state

Cal.com

Cal.com

CockroachDB

Cockroach Labs

3 in reviewView repo →
in review
Decode EncodedError once in checkTxnStatusValid#173776

Removes a duplicate decode reported at 13% of node CPU in an incident

in review
Resolve ordinals nested in a parenthesized GROUP BY tuple#173777

GROUP BY (1, x) now treats 1 as a column ordinal

in review
Preserve the column name of an aliased table-returning UDF#173779

f.out_col resolves after aliasing the function

Temporal

Temporal Technologies

2 in reviewView repo →
in review
Preserve ResourceExhausted from the search attribute provider#11777

Callers see the real throttling error, not a generic Unavailable

in review
Bound schedule-to-close on worker-deployment activities#11752

A queued attempt can no longer wait indefinitely

Ollama

Ollama

3 in reviewView repo →
in review
Don't downgrade context for requests that don't specify num_ctx#18194

Avoids reloading a model with a smaller context

in review
Support max_completion_tokens in OpenAI chat completions#18028

The limit was silently dropped and generation ran unbounded

in review

Supabase Auth

Supabase

2 in reviewView repo →
in review
Stop leaking raw internal errors in RefreshTokenGrant#2813

Raw Go error text no longer reaches API callers

in review
Recognize real Postgres unique violations#2811

Detects SQLSTATE 23505, not just one hand-raised error type