SYS_READY · BENGALURU, IN · OPEN TO WORK

Backend Dev & AI Builder.

I build distributed systems that serve global airline ops, and AI agents that make engineering teams faster. Currently shipping at Amadeus IT Group while building the future on the side.

See what I'm building GitHub → Résumé ↓
€172K
Saved / year
90
Dev-days saved
70–80×
AI token cut
★ Q3'25
Spot Award
AI_LAB · ACTIVE

What I'm Building

Flagship · Quant + AI
Systematic Momentum
Scanner
Nifty 200 positional engine · point-in-time backtested
The problem

Retail swing-trading advice is untestable noise. I wanted a rules-based, interpretable scanner that ranks the Nifty 200 every evening — and, crucially, an edge I could measure against the index instead of trust on faith.

The approach

After market close it pulls NSE bhavcopy + delivery data into SQLite, computes ~15 technical / volume indicators, and scores every stock on two horizons — 30-day momentum and 45-day trend. A batched Gemini pass tags same-day news sentiment and broker calls. The composite ranks a sector-capped Top-10 with disaster-stop and hold-and-rotate levels, emailed daily.

Architecture
NSE bhavcopy + delivery% → SQLite (idempotent upsert)
↓
~15 indicators → composite score × 2 horizons
↓
Gemini news gate ◆ delivery floor ◆ sector cap
↓
Top-10 daily email → self-grading pick ledger
How it works
  1. Ingest — NSE bhavcopy + delivery% into SQLite via idempotent upserts (killed a real score-drift bug from duplicate rows).
  2. Score — ~15 indicators (RSI / MACD / ADX / ATR / delivery) into a composite on 30d-momentum & 45d-trend horizons.
  3. News gate — Gemini tags sentiment & events; the event study found no market-neutral edge, so it's a downside-only gate, not an overfit weight.
  4. Rank — delivery-floor qualification, sector-capped Top-10, disaster-stop + hold-and-rotate levels.
  5. Prove it — point-in-time backtest (IC, decile returns) + a live ledger grading every pick vs Nifty 200.
Results · backtested vs Nifty 200
+1.14%
Alpha / pick · 45d trend
+0.54%
Alpha / pick · 30d momentum
271d
Point-in-time backtest
Index-relative alpha, market-neutral & independently reconciled (two return calcs agree to 0.0000pp). Backtest caveats, stated plainly: survivorship-biased, single regime, pre-cost.
PythonpandasSQLiteNSE data Technical AnalysisGeminiBacktestinglaunchd
Private tool · request a walkthrough →
AI Agents
Optimal Code Review
Agent Pipeline
Multi-agent system · Graphify-powered context retrieval
The problem

LLM code review gets expensive fast — feeding whole files into every agent burns tokens and dilutes signal, so reviews run slow, cost too much, and still miss cross-file context.

The approach

A GitHub PR review pipeline powered by specialised agents — Static Analyser, LLM Reviewer, Test Suggester — that coordinate via Redis and report through a Critic/Scorer. The key insight: agents query a graph.json via Graphify MCP instead of loading full files, cutting context by 85%.

Architecture
GitHub PR → Webhook → API Gateway → Redis Queue                                       ↓      Static Analysis ◆ LLM Review ◆ Test Suggester                                       ↓                   Aggregator → Critic / Scorer                                       ↓                  GitHub PR Comment
How it works
  1. Static Analyser — lint, complexity & security signals at zero LLM cost.
  2. LLM Reviewer — reasons over the graph subcontext, not raw files.
  3. Test Suggester — proposes the tests a change should ship with.
  4. Aggregator → Critic / Scorer — dedupes, ranks by confidence, drops noise.
  5. Single PR comment — one structured, prioritised review back on the pull request.
Results
85%
Token reduction
~6K
Tokens / review
5+
Specialised agents
PythonMulti-agentGraphify MCP RedisDocker ComposeLLMGitHub Webhooks
View on GitHub
Measure the edge, don't trust it
A point-in-time backtest and a self-grading ledger beat conviction. If I can't measure the alpha, I don't ship it as signal.
Less context, more signal
Graph retrieval and delivery-floor filters over brute force. Smaller, sharper inputs beat feeding everything in.
Specialised > generalised
One job per agent, one responsibility per module. Coordination beats monoliths — in AI pipelines and trading systems alike.
WHERE_I_SHIP · PRODUCTION

Day Job

Software Developer 1
Amadeus IT Group · Bengaluru
Jul 2024 — Present
★ Spot Award Q3 2025 · team of 40 ↑ €172K saved / year ↓ 70% security vulns ↓ 70–80× AI tokens

Own backend systems for airline departure control and flight management at global scale. Independently traced and fixed a critical SLA breach now saving €150K/yr, cut a further €22K/yr from log-processing overhead, and automated a manual reporting workflow that returned 90 developer-days to the team. Drove org-wide adoption of AST-based context tooling (Graphify) for a 70–80× drop in AI token usage, and added BlackDuck + Fortify gates to CI/CD that cut security vulnerabilities by 70%. Primary POC for a zero-impact quarterly data migration across teams.

JavaC++KafkaMongoDB EdifactSpring MVC/BootC++ Templates REST APIsPodmanAWSCI/CD
Software Developer Intern
Amadeus IT Group · Bengaluru
Feb 2024 — Jul 2024

Designed and built an event-driven data pipeline streaming real-time passenger & baggage data from check-in to flight management — replacing polling with live CDC. Architected on Kafka Connect + Debezium + MongoDB; containerised with Podman. Presented at internal showcase — outcomes shipped to production roadmap.

Java / JAX-RSKafka ConnectDebezium CDC MongoDBPodmanMaven
SIDE_WORK · OPEN_SOURCE

What I've Built

Python
Real-Time IoT Attack Detector
ANN detecting DDoS, port scanning, MitM & phishing on IoT devices in real-time. Includes SMS/email alerts and research paper.
TensorFlowANNIoT SecurityEdgeIIoT
Python
Financial Risk Prediction · IBM AutoAI
Corporate financial risk pipeline using IBM Watson AutoAI. Automated model selection on fraud datasets with Node-RED flows and full experiment notebooks.
IBM AutoAIWatson StudioNode-RED
NLP
Multilingual Toxic Classifier
NLP model detecting toxic comments across multiple languages, trained on multilingual datasets to generalise beyond English-only moderation.
NLPMultilingualText Classification
Python
Applied Data Science Externship
End-to-end data science project covering EDA, feature engineering, model training and evaluation across real-world datasets.
EDAFeature EngineeringML
BY_THE_NUMBERS

Impact

01
€172K
Saved per year
An SLA-breach fix (€150K) plus log-processing optimisation (€22K) — independently diagnosed and shipped to airline ops.
02
90
Developer-days saved
Automated a recurring manual reporting workflow, returning weeks of engineering capacity to the team.
03
70–80×
AI token reduction
Drove org-wide adoption of AST-based subgraph extraction (Graphify), replacing exhaustive file loading.
TECH_STACK · PRODUCTION_GRADE

Stack

pallav@palnag:~$ cat stack.config
languages
  • Java
  • C++
  • Python
  • SQL
  • JavaScript
backend
  • Spring MVC/Boot
  • JAX-RS
  • Node.js
  • C++ Templates
  • Edifact parsing
  • REST APIs · Maven
data & messaging
  • Apache Kafka
  • Kafka Connect
  • Debezium CDC
  • MongoDB
  • Redis
  • SQLite · pandas
devops & cloud
  • AWS (CCP)
  • Docker · Podman
  • CI/CD
  • BlackDuck · Fortify
  • EC2 · S3 · IAM · RDS
ai & agents
  • Multi-agent pipelines
  • Graphify MCP
  • LLM integration
  • Gemini API
  • TensorFlow · ANN
  • IBM Watson AutoAI
practices
  • Event-driven arch
  • Distributed systems
  • Reliability engineering
  • SLA analysis
  • Graph-based retrieval
  • Backtesting · IC
BACKGROUND

Education

B.Tech · 2020–2024
Vellore Institute of Technology
Computer Science Engineering
GPA 8.35 / 10 · SE, OS, DSA, Computer Architecture
Certification · Active
AWS Certified Cloud Practitioner
Amazon Web Services
Jul 2023 – Jul 2026
CONTACT · OPEN

Let's build
something.

Open to backend, distributed systems, and AI engineering roles. Bengaluru-based — open to relocation or remote.

Résumé Email LinkedIn GitHub Call