~ portfolio

Satvik Sawhney

software engineer

loading000%
defaultengineerrecruiter
builder

currently · shipping ai agents + backend infra at juspay
software engineer · juspay · ai agents + backend infra · srm grad ‘26.
résumé · pdf
see projects
Bengaluru · Delhi
satviksawhney.work@gmail.com
githublinkedin🤗 hfmedium
now — what i actually do
  • I build backends that hold up under real load.
  • I design multi-agent platforms around graphs, retries, and observability.
  • I evaluate frontier models against my own benchmarks.
01·about

I ship under load.

based · bengaluru

I'm Satvik — software engineer at Juspay, just graduated CSE (AI & ML) from SRM. At Juspay I work across AI agents, backend infrastructure, and distributed systems. The biggest piece right now is migrating our Haskell payment-routing core to Rust via a custom AI-agent transpiler pipeline — deterministic parse + semantic mapping + property-test verification, with classical fallbacks where probabilistic generation drifts.

On personal projects I own the full stack — databases, backend, frontend, auth, agentic systems (LangGraph or my own frameworks), deployed across GCP, Vercel, and AWS. Strong opinions about distributed systems, explicit invariants, and observability that surfaces the right thing instead of all the things.

Strong in algorithms, distributed systems, full-stack web, and ML, with a competitive-programming rating of 2000+ and an ICECA 2025 publication on privacy-preserving federated learning.

currently buildingHaskell → Rust transpiler via AI agents · core payment routing
satviksawhney.work@gmail.com·hindi · english
daily transactions · payment infra I ship
0M+
p99 at peak production load
<0ms
production DB shrunk · Barclays
0%
RL agent over LRU · matches a 31B teacher
+0%
02·experience

four internships, one trajectory.

deep dive01

Juspay Technologies

return offer

Software Development Engineer Intern

Aug 2025 — Present Bengaluru, India
16K IDs/ms
127 cells
  • Architected and shipped a Snowflake-style distributed ID generation system processing 16K+ IDs/ms per cell across 127 isolated cells, serving a payments platform of 300M+ daily transactions at sub-5ms p99 under peak production load.
  • Migrated a closed-loop payment routing engine to a multi-region active-active cell architecture, redesigning multi-armed bandit explore-exploit logic with per-cell Redis scoring sliding windows to preserve statistical confidence under asymmetric traffic.
  • Building a Haskell-to-Rust transpiler for the core payment routing codebase, powered by a multi-agent pipeline (parse → semantic mapping → property-test verification → human review) with a deterministic classical fallback where probabilistic generation drifts — derisking the migration of type-rich Haskell semantics to memory-safe Rust without lift-and-shift breakage.
HaskellRedisRedis StreamsPostgreSQLGrafanaCell architectureMulti-armed bandits
more about juspay
deep dive02

Barclays

return offer

Technology Summer Intern, BA4

Jun 2025 — Aug 2025 Pune, India
db size-50%
beforeafter
  • Built a Python/Flask Data Archival and Retrieval System (DARS), reducing production DB size by 50%, improving query performance by 40%, and backup/recovery speed by 30% across 50+ SQL Server tables.
  • Designed a recursive dependency analyzer using a custom topological sort (modified Kahn's algorithm) to guarantee 100% referential integrity during archival.
  • Exposed archival workflows via REST APIs with validation, structured logging, and batch execution; automated ETL reduced manual overhead by 70%.
  • Ran batch validation across 50+ tables verifying zero orphaned rows post-archival, enabling first-time compliance sign-off on the rebuilt system without rollback events.
deep dive03

Infosys

Application Development Intern

Apr 2025 — May 2025 Mysuru, India
500+ req/s
sustained · 10-person team
delivery1 wk early
+35% acc
  • Owned task breakdown and final delivery on a 10-member agile team, shipping a production-ready enterprise Flask backend sustaining 500+ simulated req/s — 1 week ahead of schedule.
  • Integrated Generative AI for summaries and personalized recommendations, improving content discovery accuracy by 35%; MVP reduced manual effort by 40%.
  • Designed REST endpoints with secure auth, optimized SQL queries with proper indexing, and structured logging — hardened the Flask backend to production-grade specification under sustained load testing.
deep dive04

Bryha Logistics

Freelance Software Developer

Aug 2024 — Dec 2024 Gurugram, India
engagement+30%
load timesub-1s
  • Designed and developed a production website with backend REST API endpoints for booking and customer-inquiry management, increasing online engagement by 30%.
  • Implemented a responsive Next.js + TypeScript + Tailwind frontend ensuring a clean UX and seamless integration with backend services.
  • Owned the full deployment pipeline on Vercel with edge caching and form validation hardening, achieving sub-second initial load on production traffic.
03·projects

Stuff I shipped instead of sleeping.

multi-agent platforms on LangGraph + RAG · a public RL benchmark beating LRU by 26% · a personal-finance OS that won’t let LLMs touch the math · plus schema-design and agri-ML tooling.

shipped
deep dive

LawyerUp

Multi-agent legal AI · LangGraph · hybrid RAG · streaming

lawyerup · /api/v1/chat· streaming
  • LangGraph-style orchestrator routing 8 specialist agents over a custom graph runtime with retries, timeouts, and checkpoint snapshots.
  • Hybrid retrieval (sqlite-vec + FTS5 + Reciprocal Rank Fusion + MMR diversity) over indexed Indian legal corpus with 384-dim MiniLM embeddings.
PythonFlaskNext.js 15+5
GitHubDemo
shipped
deep dive

SlotKeeper Fleet

1.5B Qwen warehouse agent · RL: SFT→GRPO→DPO · matches a 31B teacher

slotkeeper · fleet0.9412
P1
M1
P2
hot cold
  • RL-trained a deployable Qwen 2.5-1.5B + LoRA (r=16) manager via SFT → GRPO → DPO → Best-of-N to 0.9412 multi-task avg — +26% over LRU, matching the strongest frontier baseline at ~20× fewer parameters.
  • Built an OpenEnv-compatible multi-agent RL environment publicly deployed on HuggingFace Spaces; warehouse slotting framed as cache replacement under physical eviction cost.
PyTorchFastAPIOpenEnv+5
HuggingFaceGitHub
building
deep dive

ArchitectDB

AI-native database architect · 6 agents · MCP server

  • Multi-agent DB platform (Schema Architect, Migration Engineer, Index & Performance, Security & Governance, Schema Reviewer, NL→SQL) on a LangGraph workflow.
  • Diff-driven migration engine with forward/rollback SQL and risk classification; TypeScript MCP server exposing the platform to Cursor, Claude Desktop, and Zed.
FlaskSQLAlchemyAlembic+4
GitHub
building
deep dive

Finance OS

Personal-finance super-app · LangGraph · pgvector · Cloud Run

  • LangGraph multi-agent layer (9 agents: Cash Flow, Investment, Stocks, Tax, Debt, Goals, Insurance, Behavior, Reflection) over a deterministic computation core.
  • Every numerical claim traceable to a deterministic module via show-calculation drill-down; pgvector semantic memory; Supabase Auth + Google OAuth.
FastAPILangGraphPostgreSQL+5
published
deep dive

FarmWise

Multi-agent assistant for 126M Indian farmers · Gemini + Vertex AI

  • Multi-agent agricultural assistant on Gemini + Vertex AI supporting 6 Indian languages; capstone for Google's Generative AI Intensive Course.
  • Published Medium writeup detailing the architecture; Firestore as conversational store, Cloud Run (Docker + YAML) for deployment.
Gemini APIVertex AIGoogle Maps+3
Medium
06·stack

the tools, in one breath.

LanguagesPython
LanguagesHaskell
LanguagesJava
LanguagesC++
LanguagesTypeScript
LanguagesJavaScript
LanguagesSQL
LanguagesC
LanguagesSwift
AI / AgentsLangGraph
AI / AgentsLangChain
AI / AgentsOpenAI
AI / AgentsAnthropic
AI / AgentsGemini
AI / AgentsVertex AI
AI / AgentsMCP
AI / AgentsRAG
AI / AgentsFAISS
AI / Agentssqlite-vec
AI / Agentspgvector
AI / AgentsPyTorch
AI / AgentsLoRA
AI / AgentsGRPO
AI / AgentsDPO
AI / AgentsOpenEnv
AI / AgentsHuggingFace
BackendFlask
BackendFastAPI
BackendSpring Boot
BackendNode.js
BackendExpress
BackendREST
BackendSSE
BackendPydantic
BackendSQLAlchemy
BackendAlembic
FrontendReact
FrontendNext.js
FrontendTypeScript
FrontendTailwind
Frontendshadcn/ui
FrontendFramer Motion
FrontendGSAP
FrontendSwiftUI
FrontendStreamlit
DatabasesPostgreSQL
DatabasesMySQL
DatabasesMongoDB
DatabasesSQLite
DatabasesSQL Server
DatabasesFirestore
DatabasesSupabase
DatabasesRedis
Cloud / DevOpsGoogle Cloud Platform
Cloud / DevOpsCloud Run
Cloud / DevOpsDocker
Cloud / DevOpsKubernetes
Cloud / DevOpsVercel
Cloud / DevOpsOAuth 2.0
Cloud / DevOpsGrafana
Cloud / DevOpsJenkins
Cloud / DevOpsGit
Cloud / DevOpsKaggle
Languages
Python · Haskell · Java · C++ · TypeScript · JavaScript · SQL · C · Swift
AI / Agents
LangGraph · LangChain · OpenAI · Anthropic · Gemini · Vertex AI · MCP · RAG · FAISS · sqlite-vec · pgvector · PyTorch · LoRA · GRPO · DPO · OpenEnv · HuggingFace
Backend
Flask · FastAPI · Spring Boot · Node.js · Express · REST · SSE · Pydantic · SQLAlchemy · Alembic
Frontend
React · Next.js · TypeScript · Tailwind · shadcn/ui · Framer Motion · GSAP · SwiftUI · Streamlit
Databases
PostgreSQL · MySQL · MongoDB · SQLite · SQL Server · Firestore · Supabase · Redis
Cloud / DevOps
Google Cloud Platform · Cloud Run · Docker · Kubernetes · Vercel · OAuth 2.0 · Grafana · Jenkins · Git · Kaggle
05·shipping & solving · live · not for show

still writing code at 11pm.

github · SawhneySatvikvisit
public repos
…
last commit
…
last 12 weeksless more
leetcode · satviksawhneyvisit
problems solved
0+
cp rating
0+
leetcode · cf · cc
difficulty split
easy · 250 medium · 350 hard · 100
06·recognition

Receipts.

2026
May 2026
Graduated SRM · BTech CSE (AI & ML)
CGPA 9.8/10 · full-time at Juspay
Apr 2026
OpenEnv Hackathon R2 · SlotKeeper
RL-trained a 1.5B Qwen to match a 31B teacher · live on HuggingFace
2025
Aug 2025
Juspay · SDE intern · return offer
Bengaluru · payment routing + distributed ID gen + Haskell→Rust AI-agent transpiler
Jun–Aug 2025
Barclays · Tech Summer intern (BA4) · return offer
Pune · DARS · -50% DB size, +40% queries
May 2025
ICECA 2025 paper accepted
Federated Learning in Healthcare · co-author
Apr 2025
FarmWise · published on Medium
Google Gen AI Intensive capstone · 6 Indian languages
Apr–May 2025
Infosys · App Dev intern
Mysuru · 10-person team · 1 week ahead of schedule
2024
Jul 2024
Amazon ML Summer School '24
selected from 80,000+ candidates
Jul 2024
Apple iOS Developer Program
selected from 1,000+ applicants · 2 semesters
Jun 2024
JPMC Code for Good '24
Mumbai hackathon · built EduMate FLN platform
2023
Aug 2023
Technical Team Lead · Ramanujan Math Club
led 8-person team · platform serving 3000+ students
2022
Aug 2022
SRM Institute · BTech CSE (AI & ML)
CGPA 9.8/10 · graduating cohort 2026
2019
2019–2022
Delhi Public School R.K. Puram
Class X 96.4% · Class XII 95.6% (CBSE)
expand · full list of selections & certifications ↓
  1. 01

    Offers from Juspay & Barclays

    Aug 2025

    Return offers as SDE at Juspay and Software Engineer (BA4) at Barclays following summer internships.

  2. 02

    Apple iOS Developer Program

    Jul 2024 — Apr 2025

    Selected among 1000+ applicants for a 2-semester specialized program with Apple & Infosys.

  3. 03

    Amazon ML Summer School '24

    Jul 2024

    Cleared ML assessment; selected among 80,000+ candidates nationwide.

  4. 04

    Code for Good '24 — J.P. Morgan

    Jun 2024

    Built EduMate, an FLN assessment platform, at Mumbai hackathon.

  5. 05

    Google Generative AI Intensive & Prompt-a-thons

    2024 — 2025

    Completed Google's Gen AI Intensive (capstone = FarmWise); multiple Prompt-a-thon submissions.

  6. 06

    ICECA 2025 Publication

    2025

    Co-authored "Federated Learning and its Uses in the Modern Medical Industry" — privacy-preserving AI in healthcare.

  7. 07

    Competitive Programming

    Ongoing

    Rating 2000+ across multiple contests; 700+ problems solved on LeetCode, CodeForces, CodeChef.

07·education

where I learned things.

09·writing

Writing, on my own terms.

all writing