Lead AI Engineer · Business Analyst · Hyderabad, IN
AI SYSTEMS THAT GENERATE REVENUE. NOT DEMOS.
I'm Adithya Reddy. For 3+ years I've shipped agentic pipelines, ML models, and BI systems that survived contact with real businesses — ₹1.4Cr+ in measured impact across 4 organizations.
Scroll to initializeClaude API · LangChain · Python · SQL · Power BIPublished ML Researcher
₹1.4Cr+ Business impact3,000+ Qualified leads20+ hrs/wk Automated91.8% Ensemble accuracy40% Faster AI deployment₹8L/yr Ad spend saved5 Autonomous agent pipelines2 Peer-reviewed papers
02 · Career Evolution Engine
Promoted three times. Shipped the whole way.
From marketing analytics to owning the AI stack — every role ended with systems still running in production.
Nov 2025 — Present
Lead AI Engineer
Eject Solutions Pvt Ltd · Hyderabad
Full ownership of stack decisions, vendor evaluation, and AI integration — accelerating product deployment 40%. Designed the end-to-end v1.0 agentic architecture (tool use, memory, multi-agent orchestration) now adopted as the production blueprint across all client accounts. Co-built a market-ready AI agent with KL University, cutting outsourcing costs by ₹4L and delivery time by 80%.
LangChainClaude APIMulti-agent6-person team
Aug 2024 — Nov 2025
Business Analyst & AI Engineer
Eject Solutions Pvt Ltd · Hyderabad
Engineered LLM automation pipelines recovering 20+ hours weekly (~₹1.2L/month in team capacity). Built Power BI dashboards with 15+ KPIs informing 8+ leadership decisions. Cut developer clarification cycles from 5 days to 1 day per feature with technical specs, data models, and API contracts.
Power BI · DAXSQLTech specsC-suite reporting
Mar 2024 — Aug 2024
Freelance AI & Data Engineer
Eject Solutions Pvt Ltd · Hyderabad
Built Python + SQL lead pipelines that generated 3,000+ segmented, qualified leads — ranked the #1 outreach campaign of the year and facilitating ₹1.4Cr+ revenue within 6 months.
PythonETLLead scoring
Mar 2023 — Mar 2024
Marketing Analytics Analyst
INT360 Design Studio · Bangalore
Automated 12 hrs/week of reporting for a team of 8. Ran 15+ Meta A/B tests with significance testing — 20% CPL reduction, ₹8L/year saved across ₹50L+ managed spend. Cohort analysis in Pandas drove a 30% conversion lift.
A/B testingPandasCPL · CTR · ROAS
2019 — 2023
B.Tech, Data Science
KL University · 8.1 CGPA
Published a peer-reviewed ML paper before graduating. Coursework across ML, AI, statistics, DBMS, and data mining.
Published researcher8.1 CGPA
03 · Revenue Impact Dashboard
Measured. Not estimated.
Every number below was tracked in a dashboard I built or a P&L someone else audited.
REV_IMPACT · cumulative
₹0CR+
Revenue facilitated by the lead generation pipeline in 6 months.
PIPELINE_OUT · leads
0+
Segmented, scored, qualified leads. #1 outreach campaign of the year.
CAPACITY_RECOVERED
0HRS/WK
Manual work eliminated via LLM pipelines — ~₹1.2L/month in team capacity.
DEPLOY_VELOCITY
0%
Faster AI product deployment through full stack ownership.
MODEL_PERF · clinical ML
0ROC-AUC
Stacked XGBoost + LightGBM + CatBoost ensemble — 91.8% accuracy across 150K+ records.
COST_AVOIDED
₹0L+
₹8L/yr ad spend optimized + ₹4L outsourcing eliminated.
04 · AI Systems Laboratory
Projects don't sit in cards here. They run.
LEADGEN-01Turns raw data into ranked revenue
AGENCY-OSAutonomous brand comms · the G-BNI system
→ RESEARCHThat system, formalized & published on SSRN
LEADGEN-01STATUS: PRODUCTION · MAR 2024 — FEB 2025
Lead Generation Engine
Python + SQL pipelines that treat lead generation like an ML problem — every lead is an extracted, validated, feature-rich, scored row. Follow one batch through the pipeline (it executes live below):
01 · Extract
INGEST
Multi-source pull
Python connectors pull raw student records from admission forms, portals and sheets into one SQL store.
02 · Clean
DEDUPE
Validate · fuzzy match
Field validation plus fuzzy-match deduplication kills junk rows before they can poison the scoring model.
03 · Segment
COHORT
Features · not guesses
Leads are cohorted by course intent, geography and engagement signals — engineered features, not vibes.
04 · Score
0+
Ranked · qualified
A weighted scoring model ranks every lead by conversion readiness. Only the top tier ships to outreach.
05 · Activate
₹1.4CR+
Revenue · 6 months
Qualified leads route to campaigns with segment-matched messaging — and revenue gets attributed back.
live · pipeline.log
Why it's unique
Most agencies buy lists and spray. This pipeline engineered marketing like a data product — dedup, feature engineering, ranked scoring, revenue attribution. That discipline is why one campaign produced 3,000+ qualified leads, facilitated ₹1.4Cr+ in 6 months, and was ranked #1 outreach campaign of the year.
Ranking
#1 campaign of the year
Stack
Python · SQL · ETL
Attribution
Lead → segment → revenue
AGENCY-OSSTATUS: PRODUCTION · DEC 2025 — PRESENT
Autonomous Agency Operations
↳ Production system · formalized as EABN / G-BNI on SSRN
The autonomous system that runs a brand's communication end-to-end — five+ LangChain & Claude agents handling scheduling, content, client reporting and analytics as one stateful organization, not five chatbots. This is the live production deployment I later formalized into the G-BNI research framework (Research Lab, below). The data lanes show its coordination:
agent.scheduler● running
agent.client_reports● running
agent.content_gen● running
agent.insights● running
agent.memory_store● running
Report delivery
3 days → 4 hours
Capacity recovered
20+ hrs/wk
01 · Plan
QUEUE
Proactive · 7-day
The Topic Agent plans a rolling content + work queue ahead of time — the system never waits to be asked.
02 · Route
DECOMPOSE
Orchestrator assigns
The orchestrator splits each cycle into tasks and hands each to the right agent with the right tools.
03 · Run
PARALLEL
Shared memory
Agents generate, report and schedule concurrently against one shared state of every account and post.
04 · Deliver
4 HRS
Was 3 days
A human approves at the boundary; output ships; outcomes write back to memory and sharpen the next run.
State-aware content lifecycle · every post is a managed entity
PLANNEDTopic Agent queues it with an assigned emotional register
GENERATED5-pass narrative + schema-driven visual created
PENDINGFull preview dispatched to a human editor on Telegram
APPROVEDEditor confirms — or regenerates just one component
PUBLISHEDThree-phase Meta Graph API publication, fully audited
ANALYZEDEngagement captured at 24h · 72h · 7d, fed back to planning
↻ REGENERATING — editor kept the image, re-rolled only the caption. No restart.
Most "AI automations" are stateless one-shots; this one is a persistent, stateful organization where every agent shares memory and tool use. It runs in production across every Eject client account, collapsed report delivery from 3 days to 4 hours, and recovered 20+ hrs/week — then became rigorous enough to formalize as a peer-style research construct. The architecture below in the Research Lab is this exact system, written up.
05 · Product Architecture Universe
The v1.0 agentic blueprint. Adopted across every client account.
The architecture I designed end-to-end — now Eject Solutions' production standard. Tap any node to inspect it; data climbs up, decisions flow back down.
data climbs decisions descend
LAYER
—
—
06 · Research Lab
Two papers. Both running in production.
Research that ships: a new theoretical construct for autonomous brand AI, and a clinical-grade ensemble platform — each published, each deployed.
PAPER-01 · SSRNWORKING PAPER · MAY 2026 · ABSTRACT 6845958
Generative Brand Narrative Intelligence
The formal write-up of AGENCY-OS (in the Systems Lab above). It names a capability the field was missing — autonomous outbound brand communication — defines it with five testable properties, and presents that production system, EABN, as the first reference architecture for the construct. Where Pulsar & PeakMetrics listen, G-BNI speaks.
The five qualifying properties · P1–P5
P1 · Operational continuityRuns as a persistent service on a schedule — no per-session human invocation.
P2 · Emotional adaptivityTonal diversity enforced by architecture across 8 registers — not left to prompting luck.
P3 · State-aware lifecycleEvery artifact carries tracked states — audit trails and mid-pipeline recovery for free.
P4 · Governed autonomyHuman authority embedded at the publish boundary, with component-level intervention.
P5 · Closed-loop analyticsPublished performance flows back into future generation decisions.
Jasper ✗ · Hootsuite ✗ · HubSpot ✗ · AutoGen ✗ — EABN satisfies all five by design.
LangGraph StateGraph orchestration, Claude narrative intelligence, Telegram governance, Meta Graph publishing — containerized on Railway with health endpoints and structured audit logs.
Three-tier risk stratification that moves past binary detection — a stacked XGBoost + LightGBM + CatBoost ensemble trained across the Pima, NHANES and UK Biobank cohorts (150K+ records), with calibrated probabilities clinicians can actually trust.
Lifestyle screening → clinical assessment → full metabolic profiling, with AI-assisted recommendations conditioned on the complete risk profile. Built for clinicians, not leaderboards.
A live agent — running on the Claude API, the same stack I build with — briefed on my full track record. Ask it anything a recruiter would.
adi-twin — claude-sonnet — session active
ADI.TWIN
Online. I'm Adithya's AI twin — briefed on 3+ years of his shipped systems, metrics, and stack. Ask me about the ₹1.4Cr pipeline, the agent architecture, or whether he's worth interviewing.
>
Session active · your questions help me reach Adithya directly.
08 · System Arcade
Beat the bot. If you can.
ADI-BOT plays with a perfect minimax strategy and usually opens the game — it cannot lose, and a draw is the best you'll get. Your face is the plain marker; the bot plays as me. Manage a draw and you've earned the wall.
Your move — you're You, the bot is ADI.
W 0D 0L 0
WALL OF DRAWS · live
Nobody's held the bot to a draw yet.
Only draws (or the impossible win) make the wall. Every result pings me in real time.
09 · Mission Contact Terminal
LET'S BUILD SOMETHING THAT PAYS FOR ITSELF.
Hiring for AI engineering, analytics, or product roles — or need agentic systems built? Open a channel.
adi-twin — claude-sonnet — session active