Manas Mishra
Résumé

SDE-2 Founding Engineer Bangalore, India

Manas Mishra Backend & AI Systems Engineer

I take products from zero to one — then scale them to 100× the load. Data pipelines, distributed backends and AI agents that hold up in production.

See my work Résumé
Diagram: data from marketplaces, quick-commerce apps and private vendor portals flows through integrations, an SQS queue and parallel Lambda workers into Postgres and Redis, feeding an AI agent that produces insights. marketplaces quick-commerce vendor portals integrations sqs λ · 01 λ · 02 λ · 03 postgres redis ✦ ai agent INSIGHTS
More load handled after scaling the backend — database and services
100×
Faster cron jobs, so new clients onboard sooner
10×
Server capacity unlocked by a cross-region Postgres migration
2×
Lower LLM cost on a LangGraph node in an llm-graph-optimizer pilot
91%
  • Python
  • Django
  • FastAPI
  • PostgreSQL
  • Redis
  • MongoDB
  • AWS Lambda
  • SQS
  • Azure
  • GCP
  • LangGraph
  • OpenAI Agents SDK
  • Playwright
  • Docker
  • nginx
  • Airflow
  • Jenkins
  • Express.js

01 About

I'm happiest in the parts of a product nobody sees — the queues, migrations, pipelines and agents — making them fast, reliable and worth what they cost.

I'm a backend and AI engineer (SDE-2) with 5 years of experience — 4.5 years full-time plus a 6-month internship — building scalable, cloud-native systems in Python — Django, FastAPI and microservices on AWS, Azure and GCP. As a founding engineer at Datavio in Bangalore, I work closely with the CTO on a 0→1 data-analytics platform for brands selling on e-commerce and quick-commerce — owning architecture, deployment and post-launch optimization.

Before that, I was the technical co-founder of Bentofi, where I built an MVP that automated the manual workflows of CA firms — GST returns and reconciliation, bookkeeping and client documents — and worked with our first two clients.

Lately I've been building AI that earns its keep in production: an autonomous data-analysis agent on the OpenAI Agents SDK, agentic workflows that automate supply-chain operations, and llm-graph-optimizer, which finds the LangGraph nodes that don't need an expensive model.

Based in
Bangalore, India
Open to
Bangalore · Noida · across India
Roles
Backend Engineer · AI Engineer · Senior Software Engineer
Experience
5 years · 4.5 full-time + 6-month internship

02 Projects

Selected projects

Every project at a glance. Each one links to a full case study — the problem, the intuition, how it was built and what it changed.

$ lgo report

nodeagreep50 ms$/mo
classifyswap 98.7% 310→96 4.20→0.38

plan, respond · open-ended → keep on llm

✓ 1 node worth swapping · −91% cost · 3.2× faster

Pilot numbers for one classify node.

Side project2026AI infra

llm-graph-optimizer

Find the LangGraph nodes that don't need an expensive model.

One line instruments a compiled graph. lgo shadow replays real traffic through a lighter model, and lgo report shows per-node cost, latency and agreement — with no changes to the agent.

Cost on one classify node ($4.20 → $0.38/mo)
−91%
Faster p50 latency (310 → 96 ms)
3.2×
Agreement at confidence > 0.8
98.7%
  • Python
  • LangGraph
  • SQLite
  • CLI
  • LLM evaluation
Read the case study
captcha · sessions · proxies portal browser fetch() normalise postgres no public api playwright portal's own api per-source upsert

DatavioData engineering

Vendor-portal integrations

Reliable data from platforms that never shipped an API.

Playwright connectors log in like a person — CAPTCHA and sessions handled — then pull reports from inside the authenticated page and land clean rows in Postgres.

Platform connectors in the integrations SDK
~40
Payments reconciled to the rupee on a real remittance sample
38/38
  • Playwright
  • Python
  • PostgreSQL
  • Alembic
Read the case study

DatavioAI agents

AI data-analysis agent

Ask a question; the agent plans the analysis.

Built from scratch on the OpenAI Agents SDK: it breaks a question into steps, runs them against real datasets and explains what it found — served by FastAPI on Azure with Redis for caching and sessions.

  • OpenAI Agents SDK
  • FastAPI
  • Redis
  • Azure
Read the case study
requests nginx a91c3e2 removed janitor · every 15 min 5d02f17 draining in-flight work up to 3 h e7b4a09 live · healthy passed 60 × 2 s gate .prev → one-step rollback

DatavioDevOps

Versioned blue-green deploys

Ship any time — even while long workflows are running.

Each release boots as its own Docker Compose project, passes a health gate, and takes new traffic through a graceful nginx reload. Older versions keep draining in-flight work, then a janitor reaps them.

Drain window for in-flight workflows
3 h
Health gate before traffic moves (60 × 2 s)
~2 min
Rollback via a .prev pointer
1 step
  • Docker Compose
  • nginx
  • Bash
  • flock
  • cron
  • Azure VM
Read the case study
reconciliation · payment lookupmapped
UTR ••••4821
Transfer received₹2,41,380.52
  • INV-10421 paid in full+1,12,450.00
  • INV-10433 part 2 of 2+98,210.40
  • INV-10440 paid in full+52,118.12
  • Debit note price difference−14,902.00
  • Early-payment charge−6,233.22
  • TDS 0.1%−262.78
Adds up to₹2,41,380.52
✓ difference₹0.00

DatavioFintech

Penny-level reconciliation

Every payment explained, down to the paisa.

Channels pay brands in lumpy transfers, net of debit notes, TDS and charges. The engine traces every UTR to the invoices it settled — and every invoice to each rupee that settled it — across POs, GRNs, invoices, debit notes and payment reports.

UTRs reconciled on a first production run
88/88
Of 1,012 payments matched on another channel
99.5%
Mismatches on a 982-row validation file
0
  • Python
  • FastAPI
  • PostgreSQL
  • Azure
  • Fintech
Read the case study

Side projectAI agents

SME Agent

Question papers in, question bank out — with a human in the loop.

A LangGraph agent that parses papers, builds question payloads and syncs Drive, Excel and a question-bank API — pausing for review and resuming any time from SQLite checkpoints.

  • LangGraph
  • Python
  • SQLite
  • Google Drive API
Read the case study

DatavioData engineering

Queue-driven data pipeline

Queue it, fan it out, stop blocking the app.

Django APIs enqueue work on AWS SQS; Lambda functions process it in parallel with S3 in between — so heavy jobs never hold up user-facing requests.

Faster processing with microservices running in parallel
2×
  • AWS SQS
  • Lambda
  • S3
  • Django
Read the case study

More work

  • Cross-region Postgres migration

    Moved Azure PostgreSQL Flexible Server out of a capacity-capped US region with a read replica — unlocking 2× server capacity.

    Azure · PostgreSQL · Case study ↗

  • Document-AI reconciliation

    Debit-note PDFs → Azure Document Intelligence → SKU matching, rules first and GPT for the long tail — cutting GPT calls from 486 to 2.

    Azure AI · GPT · Python · Case study ↗

  • Async workers for slow endpoints

    Moved long-running synchronous endpoints onto background workers for more reliable APIs.

    Python · Workers · Queues

  • Market-intelligence platform

    Unified analytics tracking trends across marketplaces — scraping at scale with proxy management, browser fingerprinting and CAPTCHA handling, plus ML-driven insights.

    Django · Scrapy · PostgreSQL · MongoDB

  • Supply-chain automation

    Automated workflows built on agentic AI that take repetitive supply-chain operations off clients' plates.

    Agentic AI · Python

  • Alerts & logging service

    Real-time error tracking with Slack and email alerts, plus a centralized logging dashboard for system health.

    Django · Slack SDK · Azure

  • Shopify app

    Built and shipped a Shopify app for merchants.

    Shopify

  • Psychometric assessment platform

    AI-driven platform that evaluates candidates' psychometric and cognitive abilities.

    Flask · SvelteKit · Google Cloud

03 Experience

Work experience

5 years across a founding-engineer role, a startup I co-founded and product engineering — with a habit of owning systems end to end.

— Present

SDE-2 · Founding Engineer

Datavio · Bangalore Current

  • Work directly with the CTO in a 0→1 startup, leading backend development for data-analysis services — architecture, deployment and post-launch optimization.
  • Scaled backend infrastructure to handle 100× more load, across both the database and service layers.
  • Migrated Azure PostgreSQL Flexible Server to a new region via read-replica promotion, unlocking 2× server capacity the original US region couldn't provide.
  • Built an AI data-analysis agent from scratch on the OpenAI Agents SDK, and agentic workflows that automate supply-chain operations.
  • Built connectors for e-commerce and quick-commerce platforms — including vendor portals with no public API — saving clients hours of manual work.
  • Built a penny-level payment reconciliation engine that traces every UTR to invoices, debit notes, credit notes and TDS — 88/88 payments reconciled on a first production run and 99.5% matched on another channel.
  • Optimized cron jobs for 10× faster execution, accelerating client onboarding.
  • Designed data pipelines on AWS S3, SQS and Lambda that run microservices in parallel instead of in sequence — 2× faster processing.
  • Introduced versioned blue-green deploys on Docker Compose and nginx: health-gated releases, 3-hour drains for in-flight work and one-step rollback.
  • Designed and deployed microservices in Python, Django and Express.js on Azure, owning end-to-end API development and availability.

—

Technical Co-founder

Bentofi · Startup

  • Co-founded Bentofi to automate the manual workflows of Chartered Accountant (CA) firms, owning product engineering and building the MVP end to end.
  • Built two intake paths — Snap & Upload for photographed receipts and bills, and email integration for forwarded documents — feeding entries straight into the books behind an approval window.
  • Automated GST returns and reconciliation, client document collection and bookkeeping data entry.
  • Worked with 2 CA-firm clients on the MVP.

—

Python Developer

APlus Topper · Hyderabad

  • Built scalable applications with Django and SvelteKit, improving performance and reliability.
  • Developed REST APIs for efficient data flow and faster feature delivery.
  • Deployed on GCP App Engine (Standard) with CI/CD, reducing release time and increasing uptime.

04 Engineering log

Under the hood

The less glamorous work that keeps systems fast, cheap and online — migrations, scale-ups, pipelines and fixes.

$ git log --author="manas" --oneline

  • ops(db)Migrate Azure PostgreSQL Flexible Server across regions via read-replica promotion2× capacity
  • feat(recon)Reconcile payments to invoices, debit notes and TDS down to the paisa99.5% matched
  • perf(infra)Scale backend — database and services — for a step change in traffic100× load
  • perf(cron)Rebuild cron jobs so client onboarding stops waiting on them10× faster
  • feat(deploy)Versioned blue-green deploys with health gates, drains and rollback3 h drain
  • perf(api)Move long-running sync endpoints onto async workersreliability ↑
  • perf(db)Optimize slow PostgreSQL queries on hot backend pathslatency ↓
  • ops(db)Self-host a MongoDB server instead of paying for a managed onecost ↓
  • feat(queue)Queue-based processing with AWS SQS, Lambda and FastAPI2× faster
  • ops(ci)Scale Jenkins with Azure VM build agents
  • feat(data)Data pipelines on Python, Postgres and Airflow across Azure and GCP
  • feat(obs)Centralized logging for backend services and scheduled jobs
  • feat(scrape)Scraping at scale: proxy management, browser fingerprinting, CAPTCHA handling
  • feat(viz)Internal data dashboards on Appsmith
  • feat(app)Build and ship a Shopify app

05 Skills

Skills & stack

The tools I reach for, grouped by where they sit in the system.

Languages

  • Python
  • JavaScript
  • SQL
  • Bash

Backend & APIs

  • Django
  • FastAPI
  • Flask
  • Express.js
  • REST
  • GraphQL
  • Microservices

AI & LLMs

  • OpenAI Agents SDK
  • LangGraph
  • LLM integrations
  • Agentic workflows
  • Document AI
  • Machine learning

Data

  • PostgreSQL
  • MySQL
  • MongoDB
  • Redis
  • SQLite
  • Airflow
  • Scrapy
  • Playwright
  • Appsmith

Cloud

  • AWS S3
  • SQS
  • Lambda
  • Azure VMs
  • Azure PostgreSQL
  • GCP App Engine

DevOps

  • Docker
  • nginx
  • Jenkins
  • CI/CD
  • GitHub
  • Blue-green deploys
  • Logging & alerting

06 Education

Education & recognition

Where the fundamentals came from — an engineering degree and an international training programme — plus milestones from competitive programming and community work.

B.Tech · CSE · 2019–2023

8.98 / 10

United College of Engineering and Research

Computer Science & Engineering, Prayagraj.

International training · Thailand

1 month

Asian Institute of Technology (AIT)

Selected for a month-long international training programme at AIT, Thailand.

TechGig · 2021

Semifinal

Code Gladiators 2021

Semifinalist in one of India's largest coding competitions.

Google · 2021

#2745

Google Kick Start 2021

Ranked 2745 in Google's algorithmic programming competition.

UiPath · Community

Champion

UiPath Student Developer Champion

Led the campus developer community for a year and hosted RPA & automation events.

07 Contact

Let's build something that scales.

Backend systems, data pipelines or AI agents — if it needs to be fast, reliable and production-ready, I'd love to hear about it. Open to backend, AI and senior software engineering roles in Bangalore, Noida and across India.

LinkedIn Résumé

Bangalore, IndiaEnglish · Hindi