Work
AI Product Manager and builder based in Dubai. I shipped Vertex AI into production at Mashkor, and now most of what I build is AI agents — across conversation, research, ops automation, and decision systems — with the governance and evals that make them work in production, not just demo.
My strongest work sits at the intersection of product judgment and technical execution. I came up as a software engineer, so I build the prototypes myself: agent systems with tool-use and guardrails, recommendation systems, eval loops, and API-driven tools. I care about the full loop — user problem, model behavior, quality checks, roadmap trade-offs, launch, and measurable business impact.
60+ → 4
Agent strategies tested → promoted (>90% killed before live capital)
3×
MAU at Mashkor · shipped Vertex AI (+15% activations)
5×
Growth at PriceLabs · team 7 → 150+
AED 2M
Built 0→1 at Dash Capital in 18 months
Want the short version? Ask my AI clone anything about my experience — or read the full timeline below.
💬 Talk to my cloneIndependent · Self-directed · Dubai, UAE
Tested 60+ automated agent strategies, promoted only the 4 that cleared the rigor gates — a >90% kill rate. I build the governance that makes agents safe to run unattended.
- → Built an evaluation-and-promotion pipeline: tested 60+ automated agent strategies and promoted only the 4 that cleared statistical rigor gates — a >90% kill rate, so unvalidated logic never touches live capital
- → Systematized capital allocation across 8 parallel workstreams — a centralized layer with reservations and conflict / duplicate guards that eliminated double-allocation and the drag that was quietly eating returns
- → Operate AI agents in production with staged promotion (shadow → paper → live), tested kill-switches, human-in-the-loop, and expected-throughput watchdogs that cut silent-failure downtime to near-zero — wrote up the approach as RFC #7218 on preventing catastrophic agent actions (open-source, 8/8 tests)
- → Claude-native daily — Claude Code (hooks, slash commands, MCP servers), the Anthropic Agent SDK, tool-use, prompt engineering, LLM evals; came up as a software engineer (Java / microservices), so I build the prototypes myself
- → On the side: pavan.blog (Claude-powered conversational agent) · Angel portfolio: xAI, GrowthX, WorldMobile, Worldcoin
Founder-built venture · Dubai
Founded and grew a business 0 → AED 2M in 18 months — by building the AI ops product that ran it.
- → Built the AI-powered operations stack that ran the company — automated client onboarding (KYC/compliance-aware), buyer/seller outreach sequencing, CRM and lead-gen workflows — replacing what would normally need a 3–5 person ops team
- → Built dubai-re-intelligence (open-source): a Flask + Pandas pipeline turning raw DLD transaction data into decision dashboards that drove every allocation call
- → Full P&L ownership from zero to AED 2M (~$545K) annual revenue in 18 months — founder-operator, not a hired role
- → The takeaway: the part I loved most was building the systems — which is why I went all-in on AI product
Hyperlocal · B2C · Kuwait
MAU 3× (7K → 25K), revenue 2.8× in 18 months.
- → Led OKR strategy — roadmap focused on acquisition, retention, engagement, and revenue growth
- → Improved activation loops by 40% in 8 months via journey mapping, onboarding, and ARIA framework
- → Shipped ML recommendation engine (Google Vertex AI) — owned feasibility, eval-set design, ranking output iteration, A/B framework, +15% activations, +1.3× engagement
- → Identified SOM of 500K users, developed two core personas for "Buy Anything" and "Pick Up Anything"
- → Cross-functional leadership: UI/UX, Engineering, Marketing, Customer Support, Finance, Legal
Insurtech · B2B · India
2.5× website traffic, +30% product leads in 4 months.
- → Pioneered product-led website initiative — improved lead generation by 20% in 4 months
- → Built LinkedIn ABM campaigns that amplified product leads by 30%
- → Automated sales funnel and refined outreach — reduced response times by 25%
- → Achieved 80% OKRs for two consecutive quarters through growth experiments and process improvements
Enterprise Web Agency · Remote
+30% template discovery for 100K+ installs in 3 months.
- → Optimized Google Web Stories plugin using competitive search strategies
- → SEO consulting for HCL — measurable improvements in search visibility within 3 months
- → Upgraded digital web practices for enterprise clients — ensured effective crawling and indexing
Founder · Remote
5× growth for PriceLabs. 300K users across 6 content platforms.
- → Drove 5× growth for PriceLabs (vacation rental AI SaaS) over 16 months — 7 → 150+ people
- → Grew property listings 240% (50K → 170K) for a real estate client
- → Built content websites with affiliate marketing — 300K users across 6 platforms
- → Sold 2 content websites within a year at premium valuations
Pune, India
Business strategy, market analysis, and revenue growth.
- → Developed and executed business strategies including sales and business planning
- → Identified new opportunities and drove revenue through market analysis
- → Led business analysis to streamline operations and improve performance
Systems I've Shipped
Each system below is a product decision — what to build, what to gate, and what not to build. The domain varies, the judgment pattern is the same.
Astro · Claude API (claude-sonnet-4-6) · SSE streaming · Vercel
Conversational AI clone on pavan.blog. Designed the knowledge base, system prompt, suggested-question UX, and a graceful contact fallback for when the clone hits its limits. Streams token-by-token over SSE on a Vercel serverless function. Said no to RAG — 16 articles fit cleanly in a static knowledge base.
AlphaGrid — Production Orchestration Layer
Python · Flask · systemd · Webhook signal routing · Telegram alerts
Autonomous systems that move money need a guarded layer between the decision engine and execution — otherwise a model bug becomes a wallet bug. Built and operate a Python orchestration service routing signals from upstream decision systems through a risk guardian (drawdown-kill, per-strategy loss caps, conflict and duplicate guards, two-stage entry, tested kill switch) and staged-promotion gates (shadow → paper → live). Telegram alerts on every entry, close, and error. Said no to hooking every upstream system immediately — only ones that pass the pre-production gate are enabled. Public pattern extracted to GitHub; live dashboard and production adapters remain private.
Lab Framework — Multi-Stream Promotion Infrastructure
Python · Flask · cron · Plug-in stream registry
Once you scale beyond two production streams, ad-hoc promotion decisions become the bottleneck — and the source of every avoidable incident. Built a Lab Framework where each candidate stream registers its own gate criteria (statistical thresholds, capital limits, error tolerances), and a nightly review cron measures every stream against its criteria. Two endpoints surface the state of the world: /api/live-readiness reports which streams have passed all gates, /api/risk-status reports which need attention. Said no to manual promotion overrides — every promotion is gate-driven and audit-logged.
Dubai RE Intelligence
Python · Flask · Pandas · DLD transaction pipeline
Dubai real-estate decisions at Dash Capital were being made against scattered DLD exports and manually-pulled data — slow and hard to re-run. Built a Flask + Pandas toolkit that auto-loads DLD transactions, normalises two incompatible export formats, and focuses the view on the two communities driving the firm's thesis: Emaar South and Dubai Creek Harbour. Said no to a generic all-of-Dubai view — focused on communities that drive decisions, not vanity breadth.
Content Research Agent
Python · Anthropic Claude API · Cron-friendly
Content research is the slowest, most expensive part of running a niche newsletter when done by hand. Built two Claude-powered agents that turn a Monday morning's research into a 2-minute cron job: one agent surfaces trending topics, pain points, and regulatory updates (VARA, UAE Central Bank); the other runs a YouTube content-strategy brief with hook titles and content gaps. Said no to RAG, scraping, and vector DBs — a single structured prompt is enough for weekly cadence. Keep the complexity budget for downstream production.
Angel Investing
Education & Certifications
Gen AI Product Strategy
Walmart AI Leaders · 2024
Gen AI — Idea to MVP
Uber AI Leaders · 2024
Product Strategy
Reforge · 2023
Master in Product Management
Reforge · 2022
Product & Growth Bootcamp
GrowthX · 2021
Advanced Google Analytics
Advanced SEO Strategies
UC Davis
Content Marketing
HubSpot Academy
Education
Bachelor of Engineering (BE) in Information Technology — University of Pune, 2006
Tools & Skills
AI & Agents
Product & Growth
Analytics & Tools