Skip to content

SS.SYS— Soumya Shekhar, back to top

[ 00 · The software story ] Backend, Systems & Applied AI Engineer

Soumya Shekhar — Code that
thinks fast.

Portrait of Soumya Shekhar
Video telemetry latency
sub-100ms
Outbound email engine
10K+ / day
Rule-engine accuracy
98.89%
3x hackathon champion
1st Place
Live · Neural mesh

I'm Soumya Shekhar — I build backends that answer in milliseconds, pipelines that move thousands of messages a day, and AI agents that check their own answers. Software engineering at DTU.

Switch personality

Same person, two stories. Flip it to meet the drone founder.

Read the build logGet in touch

Chapter 01 · Mindset

How I engineer systems.

I believe great software engineering balances low-level control with high-level AI capabilities. Whether optimizing Ethernet video transport for sub-100ms telemetry or framing stateful multi-agent RAG pipelines, I focus on predictable, sub-second latency and resilient architecture.

My work spans distributed email delivery engines (10K+ msg/day), async vehicle damage estimation platforms, explainable deterministic rule engines, and autonomous hardware telemetry.

soumya@dtu-workstation:~/architecture
// Core Engineering Directivesconst ARCHITECTURE_TENETS = {  telemetry: "Sub-100ms video transmission & state updates",  events: "500+ events/min with structured capture & alerting",  delivery: "10K+ messages/day with 98.5% delivery success",  multiAgent: "Stateful LangGraph DAG with MMR reranking & validation",  determinism: "28-attribute rules, 98.89% accuracy, zero network dependency"}; console.log("Ready for deployment & collaboration.");

Chapter 02 · Journey

Two internships, one obsession with latency.

Event streams and computer vision first, then real-time video over the wire — told in the order it happened.

  1. Part 1

    Jun 2025 — Aug 2025

    New Delhi, India

    HOUSE TECHNOLOGIES

    Backend & ML Engineer Intern

    Architected REST services, high-throughput event processing pipelines, and integrated computer vision object tracking.

    • Architected REST APIs and video-processing pipelines integrating ML-based tracking for real-time object detection and analytics.
    • Provisioned telemetry services processing 500+ events/min with structured event capture and alerting, slashing manual debugging overhead by 30% and accelerating incident response.
    • REST APIs
    • Python
    • FastAPI
    • Computer Vision
    • Telemetry Services
    • Event Alerting

  2. Part 2

    Jun 2026 — Aug 2026

    New Delhi, India

    HEURONICS

    Software Engineer Intern

    Moved from event streams to video streams: a transport and encoding pipeline fast enough for an operator to steer by.

    • Constructed a real-time telemetry and control platform for sensor data, mission state, and remote AUV/ROV operation over local Ethernet networks.
    • Optimized a low-latency video transmission pipeline through refined encoding and transport protocols, achieving sub-100ms latency for dependable remote monitoring.
    • C++
    • Python
    • Networking
    • Ethernet Protocols
    • Real-Time Telemetry
    • Video Encoding

Chapter 03 · Venture

About Vektordynamics.

Vektordynamics is a deep-tech robotics company in the autonomy segment, building autonomous aerial, ground and marine systems that perceive, navigate and decide on board.

vektordynamics.in (opens in a new tab)

Vektordynamics Private Limited

Autonomy for aerial, ground and marine systems

Founded
2026
Headquarters
New Delhi, India
Founder
Soumya Shekhar
Domains
Aerial · Ground · Marine

Recognition

  • Top 60 startups, Delhi Startup Yuva Festival 2026
  • Featured by NDTV
NDTV coverage (opens in a new tab)

Its systems run AI on board, link to the ground over hybrid radio channels and keep operating without cloud or network dependency. The aircraft, ground-station software and companion apps are developed as one stack, with in-house BLDC motors and encrypted radio links.

What the company builds

  • JATAYU

    Autonomous UAV with Onboard Edge AI

  • JATAYU FIELD COMMAND

    Multi-Drone Ground Station & Mission Control

  • SOS GUARDIAN

    Fall Detection & Emergency Alert App

  • VEKTOR PLATFORMS

    Drones, Links, Docks & Motors

Services

  • Drone procurement
  • Field pilots
  • Contract R&D
  • Thermal inspection

Chapter 04 · Builds

Systems shipped end to end.

Distributed workers, agent graphs and offline inference — each one measured, not just described.

Build 01 · Distributed Systems & Outreach

MAILALL

Distributed Outbound Email Platform

Throughput
10K+ / day
Delivery Success
98.5%
Discovery Speedup
60%
Workers
Distributed

Engineered a distributed outbound-email platform with background workers for AI-personalized campaigns, processing 10K+ emails/day with 98.5% delivery success rate and intelligent batch delivery with reply tracking.

Developed a custom mail-sending and enrichment pipeline using Nodemailer, IMAP, and Scrapy, slashing contact-discovery time by 60% through contextual AI-powered enrichment.

  • Next.js
  • Node.js
  • Express.js
  • LibSQL
  • Kysely
  • Nodemailer
  • Scrapy
GitHub — MAILALL (opens in a new tab)

Build 02 · Multi-Agent Systems & GenAI

TECH-RESEARCH AGENT

Multi-Agent RAG Platform

Live Sources
12+
Aggregation Time
< 15s
Answer Relevance
97.8%
Crawl Success
99.1%

Orchestrated a stateful multi-agent system using LangGraph DAG orchestration, aggregating research results across 12+ live sources in under 15s with 97.8% answer relevance.

Refined MMR + LLM reranking across 100 candidate chunks with iterative self-correction, achieving 98.2–99.1% crawl success rate.

  • Python
  • LangGraph
  • Qdrant
  • Groq
  • Crawl4AI
  • RAG
  • LLM Reranking
GitHub — TECH-RESEARCH AGENT (opens in a new tab)

Build 03 · Explainable AI & Edge Systems

Allele 7.o

Offline Explainable Decision Support System

Accuracy
98.89%
ROC-AUC
0.998
Network Dependency
0%
Rule Attributes
28

Implemented a 28-attribute deterministic rule engine, achieving 98.89% accuracy, 0.983 recall, and 0.998 ROC-AUC on the evaluation set.

Deployed an explainable AI agent into a cross-platform Electron application, enabling sub-second local inference with zero network dependency and complete decision transparency.

  • Electron
  • React
  • FastAPI
  • SQLite
  • Local Inference
  • Python
GitHub — Allele 7.o (opens in a new tab)

Chapter 05 · Toolkit

The instruments behind the work.

Languages, infrastructure and frameworks that show up again and again across the builds above.

Showing 6 categories

Languages

07
  • Python
  • C++
  • Java
  • Rust
  • TypeScript
  • JavaScript
  • SQL

Backend & Systems

08
  • FastAPI
  • Node.js
  • Express.js
  • REST
  • GraphQL
  • Docker
  • Redis
  • Linux

Data & Infrastructure

07
  • PostgreSQL
  • SQLite
  • MongoDB
  • Supabase
  • Celery
  • WebSockets
  • AWS S3

AI / ML & Frameworks

07
  • PyTorch
  • LangGraph
  • RAG
  • LLM Integration
  • Qdrant
  • Crawl4AI
  • Scrapy

Engineering Tools & QA

07
  • Git
  • GitHub
  • Semgrep
  • SonarQube
  • Kysely
  • LibSQL
  • Nodemailer

Core Competencies

10
  • Backend Systems Design
  • Distributed Systems
  • REST & GraphQL APIs
  • Asynchronous Processing
  • Real-Time Data Pipelines
  • Microservices
  • Relational & NoSQL Databases
  • Message Queues
  • System Reliability
  • Applied Machine Learning

Chapter 06 · Milestones

Where it was learned, and where it was tested.

[ Academic Foundation ]

Delhi Technological University (DTU)

B.Tech — Software Engineering

Aug 2023 — Jul 2027New Delhi, India

[ Hackathons & Recognitions ]

1st Place Winner

Vihaan 8.0 • MLH Semicolon • Onchain Island

2025

Delivered high-concurrency full-stack and blockchain prototypes under strict 24–36 hour hackathon sprint constraints.

Chapter 07 · Epilogue

Get in touch.

For engineering roles, collaborations or questions about the work above, send a message below or view the full resume.

Official Curriculum Vitae

The complete, printable version of Soumya Shekhar's resume — formatted for engineering recruiters, ATS screeners and systems hiring teams.

Direct Contact

shekharsoumya34@gmail.com
soumyas.me (opens in a new tab)

Network Handles

GitHub (SS-S3) (opens in a new tab)LinkedIn (soumya-shekhar) (opens in a new tab)