We ask institutions to trust records. Here is ours.
Sigilith is the authorisation layer for AI decisions that create risk. Every claim on this page carries a reference: a named venue, a named institution, a named employer.
Logs aren't evidence.
AI now produces language and decisions that can trigger scrutiny: audits, disputes, investigations.
Months later, teams are forced to reconstruct what happened. That reconstruction depends on brittle artefacts: screenshots, ad-hoc exports, tribal knowledge.
Sigilith exists to make reconstruction unnecessary: the record was already made.
The name is the specification: sigil, a seal of authority, and lith, stone.
- 01AI became operationalAutomated decisions now carry institutional risk.
- 02Governance became enforcedRegulators require proof of process, not just outcomes.
- 03Proof became mandatoryNarratives no longer satisfy audit requirements.
Minimal claims. Maximal integrity.
Artifacts over assertions
If it can't be produced later, it can't be defended.
Context belongs with the output
Policy state and approvals matter as much as the text.
Custody without ceremony
Chain-of-custody should be native, not manual.
Export is the test
When scrutiny hits, you need a packet, not a dashboard.
Three records, open for inspection.
Sigilith was founded by Syed Tasdid Azam Dhrubo, S M Jishanul Islam and Sadia Ahmmed. Research that ships, products that survived production, and operations that have answered to regulators. Select a dossier for the full record.
Syed Tasdid Azam Dhrubo
Runs an AI platform in production inside a Canadian telecom, and is building Sigilith around the governance discipline that kept it there.
Syed leads Sigilith's product direction, AI governance, and commercial strategy. He is concurrently AI Product Manager at an Indigenous-owned AI organisation in Canada, where he took an AI platform from inception into production for TELUS, now in daily use by more than a thousand employees at the telecom.
He previously led Google Maps transit-data processing across Bangladesh, and built an EdTech platform for the University of Alberta's Pharmacology department.
His discipline is the one this market demands, AI product management and AI governance: shipping systems whose decisions an institution can still defend months later. That requirement is the thesis Sigilith is built on, and it governs both the platform and Sentry.
- TELUSAI platform taken from inception to production; 1,000+ employees· Canada
- Google MapsLed transit-data processing across Bangladesh
- University of AlbertaBSc Computing Science & Psychology
- UAlberta PharmacologyBuilt departmental EdTech platform
- NeurIPS 2024MIMIC: multimodal Islamophobic meme identification (MusIML workshop)
- ICCV 2025Tri-modal adversarial attacks on short videos (SVU workshop)
- Alberta Machine Intelligence InstituteCTO coaching programme alumnus; one of Canada's national AI institutes
- MIT SolveSelected participant· 2024
S M Jishanul Islam
Architects both products end to end: the record formats and verification paths the platform rests on, and the liveness engine Sentry runs on.
Jishan owns Sigilith's engineering: the deterministic record formats, the sealing and verification paths, and the distributed infrastructure both products depend on. He has architected and delivered AI systems for the Government of Bangladesh and TELUS International, leading engineering teams from concept through deployment.
His published research spans EACL, ICCV, NeurIPS and IEEE venues, covering multimodal learning, computer vision, OCR, and verifiable-record systems. That range maps onto the company directly: provenance and integrity primitives for the platform, liveness and capture forensics for Sentry.
The overlap is not incidental. BornoDrishti, his Bangla OCR work published at EACL, addresses the extraction problem on Sentry's document-intelligence research roadmap: diverse, degraded, non-Latin scripts that defeat template-based systems.
He builds for reliability first: production systems held to institutional standards of performance, security, and scale.
- EACL 2026BornoDrishti: vision encoders & domain-adaptive learning for Bangla OCR
- Knowledge-Based SystemsAudio-video multimodal fusion for emotion recognition (Elsevier)· 2026
- ICCV 2025Tri-modal adversarial attacks on short videos (SVU workshop)
- IEEE Big Data 2025Position-aware metric & lightweight LLM for Bangla punctuation restoration
- NeurIPS 2024MIMIC: multimodal Islamophobic meme identification (MusIML workshop)
- TENCON 2025IPBlocks: verifiable registration and provenance on-chain
- ICEEIT 2024E-MedViTR: vision transformers with registers for biomedical imaging
- Government of BangladeshArchitected and delivered national AI systems
- TELUS InternationalLed engineering from concept to deployment
- United International UniversityFaculty, Computer Science & Engineering
Sadia Ahmmed
Runs the commercial and operational path into regulated buyers, and researches whether the systems actually work for the people who must use them.
Sadia directs Sigilith's commercial strategy and operations: taking both the platform and Sentry into regulated institutions, and making deployments survive procurement, compliance review, and regulatory scrutiny.
Her research spans machine learning, computer vision and human-computer interaction (how systems behave across different populations and operating contexts), published at ICCV and NeurIPS workshops, with first authorship on a vision-transformer paper at ICEEIT. It bears directly on both products: whether a governance tool is genuinely usable by the compliance officers accountable for it, and whether verification holds for an entire population rather than its most convenient users.
She has led a 3,000-member technology organisation and directed national programmes with Grameenphone, Robi, and Therap, working directly with regulators, corporate leadership, and public-sector stakeholders across Bangladesh in both English and Bengali.
She graduated Summa Cum Laude from United International University, served as a Lecturer there, and is a Master's student and researcher at the University of British Columbia.
- University of British ColumbiaMaster's student: machine learning, computer vision, HCI· current
- United International UniversitySumma Cum Laude, CSE (Data Science); later Lecturer
- ICCV 2025Tri-modal adversarial attacks on short videos (SVU workshop)
- NeurIPS 2024MIMIC: multimodal Islamophobic meme identification (MusIML workshop)
- ICEEIT 2024E-MedViTR: first author; vision transformers with registers for biomedical imaging
- arXiv 2024Bengali regional-dialect transcription to IPA with district-guided tokens
- Grameenphone · Robi · TherapDirected national programmes in partnership
- Public sector · BangladeshRegulator and stakeholder engagement, English & Bengali
- Technology organisationLed a 3,000-member body
- Hult Prize SummitTop-15 global placement, Boston
- International Blockchain OlympiadGold Award with funding
Credentials, not adjectives.
Peer-reviewed research at EACL, ICCV, NeurIPS and IEEE venues, alongside delivery inside telecoms, national governments, and universities.
This is not three separate CVs. All three founders appear as co-authors on the same peer-reviewed work: a team that has already carried joint research through review.
- TELUS
- Google Maps
- Government of Bangladesh
- University of Alberta
- University of British Columbia
- United International University
- Alberta Machine Intelligence Institute
- MIT Solve
- Grameenphone
- Robi
- Therap
Evidence, not narrative.
We work with a small number of design partners in accountability-heavy domains. If you own regulated communications, governance gates, or model risk oversight, we should talk.
- High-stakes AI outputs with real audit exposure
- Teams accountable for risk and compliance outcomes
- A credible pathway to deployment