PRODUCT LEADER · AI PRODUCTS
Prashanth
Nimmagadda
For 8 years I’ve worked at both ends of product: with telecom operators in six countries, where one release reached millions, and in my own startups, where I was most of the team. Now I bring both to AI products, from the first spec to the evaluation.
Every role, degree, build and project.
Newest first. Open a pass for what I did and the numbers behind it, or ask my AI. The numbered ones have a full chapter below.
- Work
- Study
- Build
- Project
BUILD, chapter 1 below2026RECALLSolo build · product, spec and code
AI decision intelligence for meetings. Decisions are extracted on the phone, so recordings stay private and inference costs nothing. iOS app in App Store review; Android in public testing.
- 78.9%extraction recall, restored after the iOS 27 change
- 50beta testers
STUDYSep 2025 – Sep 2026University of BirminghamMSc Business Analytics
Completed in September 2026.
WORKSep 2026University of BirminghamWelcome Registration Assistant · part-time
Registrations, record checks and campus ID issuing inside the University’s Welcome operation.
- 20,000student interactions in the operation
BUILD, chapter 3 belowMay – Sep 2026PRD Genie · open sourceRebuilt from scratch, MIT licence
A local-first PRD workbench. Bring your own model; every AI edit arrives as a cited proposal that a person has to apply.
- 234automated tests
- MITopen-source licence
PROJECT2026Fair Credit Intelligence AgentIBM SkillsBuild AI Experiential Learning Lab
A multi-agent system on IBM watsonx.ai that assesses credit-invisible applicants, with a full audit trail.
- IBM · strong finalist
PROJECTJun 2026Department for EducationFuture Leaders Certificate · consultancy project
Designed a non-grant commercial model to put solar on England’s schools, backed by a 25-year PPA financial model and presented to DfE stakeholders.
- 22,000+schools in the estate
- 5,000Monte Carlo simulations
BUILD, chapter 3 belowApr – Sep 2025PRD GenieInternal tool at Digitral
An internal tool that drafts product requirement documents, with hybrid retrieval: keywords for exact terms, embeddings for meaning.
- −70%documentation time
- 80%repeat usage in the pilot
WORK, chapter 2 belowFeb 2022 – Sep 2025DigitralProduct Manager
Products for six telecom operators, roadmap ownership for a major telco client’s consumer platform, and a field workforce platform. I led two product managers and a product analyst.
- 23M → 50MMAU, growth I contributed to
- +35%engagement after one relaunch
BUILDJan 2024 – Mar 2025Synopse AICo-founder and product lead · part-time
A personalised news platform on a RAG pipeline. I owned product, UX, go-to-market and growth.
- 5,000+users in 25 countries
- 42%weekly retention
WORKJan 2021 – Feb 2022Arrive DeliveryFounder and product lead
Consumer app, rider app and merchant portal from zero to MVP, with real-time dispatch and order batching.
- Startup India certified
- T-Hub Lab32
- Razorpay Rize · 5% acceptance
WORKAug 2019 – Jan 2021PicxyBusiness development and product strategy
A Techstars-backed martech startup. I led the market analysis behind its pivot to enterprise, which contributed to its Series B.
- 30%month-on-month revenue growth
- £2Menterprise pipeline
WORKJun 2017 – Nov 2018Government of Andhra PradeshConsultant
Managed the UT Austin-XLr8AP Technology Accelerator and drafted the state’s first startup policy framework.
- 50+startups
- 5,000students through 30 IoT bootcamps
STUDYJun 2015 – Dec 2016Texas A&M University-CommerceMSc Technology Management
STUDY2008 – 2013SRM University, ChennaiBTech Mechanical Engineering
The model changed under me. One field brought it back.
RECALL pulls the decisions out of meetings on the phone, so recordings never leave the device and inference costs nothing. In September, Apple’s iOS 27 model change broke my prompts with no change to my code. I measured the damage, reworded one field and brought recall back to 78.9%.
10,000+ agents. A 300-metre circle.
One platform to recruit, check in, approve and pay a field workforce across Indonesia’s three time zones. I went after the two known pains: slow recruitment, with AI CV screening, and attendance fraud, with location validation.
- RecruitAI CV screening
- Check in300 m geo-validation, GPS-spoofing checks
- ApproveMulti-level approvals, payroll cut-off
- PayKPI-weighted commissions, payout reconciliation, AML-linked identity checks
Two kinds of retrieval. One person decides.
Product managers asked it two kinds of question: conceptual ones, and exact terms. So the retriever does both. In 2026 I rebuilt it in the open, and the rule got stricter: every AI edit is a cited proposal, and nothing changes until a person applies it.
- Question
- KeywordsBM25 ranking, for exact termsMeaningembeddings; falls back to keywords if the model can’t load
- Cited excerptsup to 8 per action
- Modelyour key: OpenAI, Anthropic, Gemini, OpenAI-compatible, or local Ollama
- Proposal → Applya person accepts or dismisses it
2025 · INTERNAL TOOL, DIGITRAL
2026 · OPEN SOURCE, MIT · v0.1.0-rc.3
FROM MY TEST LOG Local 4B and 9B models wrote reviews with wrong citations. Validation rejected them, and nothing was applied. The 70% and 80% figures come from the 2025 internal pilot; the open-source release is new and has no usage figures yet.
Ask the rest.
My AI answers only from my verified record and shows its sources. Or print the CV. No sign-up needed.
AVAILABLE NOW · ROLES ACROSS THE UK AND EU · LINKEDIN ↗ · GITHUB ↗
Weather: Open-Meteo.com, CC BY 4.0
Anonymous visit counts via Vercel Web Analytics. No cookies.